Skip to main content
AI connector

The CheckList: a client’s month-end checklist in one request

Type “Run Checklist for Acme, Month 10-26” in Claude, ChatGPT or Grok, and your AI runs that client’s month as eight steps: bank statements, categorizing, rules, questions to your client, the QuickBooks entries to fix, the reconciliation, 1099s and W-9s, and the final checks before posting. It shows how much of the month AI did and how much is left, sorts each transaction into done by AI, matches past bookkeeping, AI’s best guess or only the client knows, and ends with one list of changes for your yes. It replaces Run the books for a client. In Claude’s + menu it is Run the CheckList for a client: if you still see the old name, reconnect LedgerSync once. See The CheckList.
  • A long list cut to fit one answer now comes with a next_cursor that starts at the first row it left off, so your AI reads every row once: the 1099 list, the review reports, the statement comparison’s lists and the history of AI changes. The W-9 request list still gives a smaller limit to call again with, and that call now fits.
Checks

The checks list comes newest first, and pages faster

GET /v3/accounts/{account_id}/checks now lists checks newest first: the check LedgerSync captured most recently is first. Before, check ids were compared as text, so on an account whose check ids differ in length an older check could come before newer ones. Cursors you already hold stay valid.
  • Each page now reads only its own checks, so a large account pages much faster: a page of 100 checks on a 548-check account took about 45 seconds, and about 1 second after this change.
AI connector

Five tools are now part of others

The live connector has 66 tools instead of 71 (sandbox 64 instead of 69): five tools were folded into the tools they always followed, so AI apps with a cap on how many tools they load keep every LedgerSync tool.
  • ledgersync_get_operation_status now reports any background work by the id its tool returned: a refresh (operation_id with connection_id, as before), a QuickBooks push (push_id), changes made in QuickBooks (change_set_id) or a rules setup (setup_id). It replaces ledgersync_get_quickbooks_push_status, ledgersync_get_ledger_changes_status and ledgersync_get_rules_setup_status, with the same answers.
  • ledgersync_list_rules with a rule_id shows one rule in full. It replaces ledgersync_get_rule.
  • ledgersync_get_statement_conversion with a format adds a one-time link to the converted file once the conversion is completed. It replaces ledgersync_get_conversion_download_link.
  • An AI app still holding its old tool list gets “unknown tool” from the five old names: reload the tool list (see Tool updates on For developers).
AI connector

Your AI remembers each client

Tell your AI something about a client, like “Acme uses Stripe for card payments and its year ends in June”, and after you confirm the words, LedgerSync keeps it. Every chat, everyone at your firm who works on the client, and every AI app you use starts from it, so the AI stops asking what it already knows.
  • New tool ledgersync_save_client_fact: saves, changes or removes one fact about a client. A client keeps at most 20 facts, each up to 500 characters and 6,000 in all; numbers of 8 or more digits (tax IDs, account and card numbers) and sign-in details are refused.
  • ledgersync_get_client_snapshot now returns the client’s facts by category, with facts_total, and for a client with none can offer to set them up.
  • Your AI app loads the new tool with the tool list (see Tool updates on For developers).
Transactions

merchant_name names the other party, plus counterparties and payment details

merchant_name now names the other party of a transaction: the merchant on a purchase, or the company or person paid or paying on an ACH or wire. It is no longer a transaction-type label such as Outgoing Wire or Orig Co Name, and three new fields describe that party and the payment.
  • Finicity names are read from the ACH and wire text. They keep the bank’s spelling and cuts: CIBC, for example, cuts an ACH company name to 16 characters and a wire beneficiary to 18.
  • MX transactions carry the name MX cleaned. Their merchant_name was always null before.
  • null when the bank text names no outside party: checks with no payee, deposits, transfers between the holder’s own accounts, loan moves, fees, interest and sweeps.
  • New merchant_entity_id: MX’s merchant ID (MCH-...), the same across banks and accounts. It is null on Finicity, portal (FDE) and uploaded-statement (PDF) transactions, and when MX’s ID is a catch-all the bank uses for many unrelated parties.
  • New counterparties: always an array. Today it holds at most one entry, the party merchant_name names, and it is empty when merchant_name is null. Each entry has name, type, entity_id and confidence_level, which is HIGH for an MX name backed by its merchant ID and LOW for a name cleaned or parsed from the bank text.
  • New payment_meta: always an object with payment_method (ACH or WIRE today), payer (money in), payee (money out) and reference_number (the bank’s reference as printed). All four are null unless the row is an ACH or wire LedgerSync can read.
An outgoing wire read from Finicity at CIBC, with the beneficiary cut to 18 characters:
The new fields are additive, and a transaction you already pulled shows the new values when you read it again. If you match on merchant_name, expect labels such as Outgoing Wire to give way to a name or null. Every field and value is in The transaction object.
AI connector

Ask your AI what it changed

LedgerSync keeps a history of every change the AI makes through it, with who approved it and from which AI app. Ask “What did the AI change this month?” or “Who approved the rules for Acme?”: your AI answers from the history and, if you ask, undoes a change that can still be undone.
  • New tool ledgersync_list_audit_log: the history, newest first, by client, dates, tool or person. Your firm’s owner and admins see the whole firm; anyone else sees their own clients and the changes they approved.
  • The firm’s owner and admins also see it on the AI changes page in the LedgerSync app.
  • Your AI app loads the new tool with the tool list (see Tool updates on For developers).
AI connector

Run the books tells you when it is waiting for you

If Claude stops a run to wait for you to allow a step, your AI now tells you what to click and goes on after your OK, instead of saying the run could not start. When you say yes to refreshing the bank connections, it also tells you each refresh may ask for your OK. No reconnect needed. See Run the books for a client.
AI connector

Just ask your AI to run the books

Type “Run the books for Acme for October” in Claude, ChatGPT or Grok, and your AI follows LedgerSync’s steps for it, with no menu needed. The run now asks for the client and the month first, checks for missing statements (a bank that never sends statements is flagged as that, not as missing), checks every categorization, then the rules, the reconciliation and the rest, and ends with one list of everything to fix, each with its solution.
  • New tool ledgersync_get_playbook: any LedgerSync playbook’s steps by name, the same as its prompt.
  • Your AI app loads the new tool with the tool list (see Tool updates on For developers). See Run the books for a client.
AI connector

Run a client’s books with one request

A new prompt, Run the books for a client, has your AI do one client’s bookkeeping end to end: the bank feeds, categorizing and posting, rules, questions to your client, the statements against the books, the month’s review and, in the 1099 season, 1099s and W-9s. It asks its questions first and shows every change in one list: nothing changes until you say yes. In Claude, pick it from the + menu; in ChatGPT or Grok, paste the instructions. See Run the books for a client.
AI connector

Every firm can now send W-9 requests from their AI

After your yes, your AI emails your client’s vendors a secure link to fill in and sign their W-9. This was open to a few test firms; it is now open to every firm. No reconnect needed. See Ask vendors for their W-9.
Clients

Updating a client keeps it in its group

PATCH /v3/clients/{client_id} used to take the client out of the client group it was filed under in the LedgerSync app, on every update, even one that only set external_id. The client now stays in its group, as every field you leave out stays as it is.
  • Nothing to change on your side: the request and the answer are the same as before.
AI connector

Ask vendors for their W-9 from Claude or ChatGPT

Ask your AI “Send W-9 requests to Acme’s vendors with no tax ID on file.”. After your yes, LedgerSync emails each vendor, from its own address under your firm’s name, a secure link to fill in and sign the official Form W-9. Ask later who has finished, and get each finished W-9 as a PDF through a link that works once.
  • Four tools, now on live as well as sandbox: ledgersync_preview_w9_requests, ledgersync_send_w9_requests, ledgersync_list_w9_requests and ledgersync_get_w9_download_link. Sending opens to a few test firms first.
  • On live, the 1099 list’s w9_status now shows each vendor’s W-9 request, and ledgersync_get_client_snapshot returns open_w9_requests. The 1099 season prompt offers the W-9 requests too.
  • Optionally, LedgerSync saves each W-9 to QuickBooks once the vendor signs: the 1099 mark, and the tax ID unless QuickBooks has a different one.
  • Tax IDs never reach the chat: the vendor enters theirs on LedgerSync’s secure page.
  • Your AI app loads the new tools with the tool list (see Tool updates on For developers). See Ask vendors for their W-9.
ConnectionsAI connector

A failed statement pull no longer reads disconnected

GET /v3/connections and GET /v3/connections/{id} used to report a Finicity connection as disconnected, with no last_error, whenever its last statement pull had failed, even while its transactions and balances refreshed every day. Some banks never serve statements through Finicity (CIBC U.S. Business is one), so their connections stayed disconnected, and reauthorizing could not help: Finicity answers “No action required”.
  • The status now follows the bank login, so those connections read active.
  • The failed pull still shows on realized_capabilities.statements, as failed.
  • Nothing to change on your side. If you skip disconnected connections, expect some of them to come back as active.
  • The AI connector no longer suggests a fix link for them.
AI connector

Clearer counts, and a summary to send after a comparison

In ledgersync_run_diagnostic’s answer, over_90_count is now over_90_customer_count on receivables and over_90_vendor_count on payables, so the count of who is owed over 90 days says what it counts.
  • When ledgersync_get_financials compares two periods, its next_step now asks the AI to end with a short summary the user can send the client, with or without the flux_and_summary prompt.
  • No new tools, and nothing to reload: the answers carry the change.
AI connector

The new-client check says who is behind old bills and invoices

ledgersync_run_diagnostic’s receivables and payables now carry evidence in words, for example “7,440.14 over 90 days: 7,925.94 owed to 2 vendors, less 4 old credits of 485.80”: over_90_count counts the vendors or customers behind the amount, not bills or invoices.
  • ledgersync_get_financials, ledgersync_drill_account and ledgersync_run_diagnostic keep each answer’s whole text under 15,000 characters, the sentence before the JSON and the client’s name included.
  • No new tools, and nothing to reload: the answers carry the change.
AI connector

Get check images from Claude or ChatGPT

Ask your AI “Get Acme’s October check images, named by payee and date.”. It lists each check (its date, number, amount, payee and memo as read from the check, and the bank transaction LedgerSync matched it to, when there is one) and gives you a link that downloads the images: one check as its image, more as a ZIP with a CSV of their details.
  • Two new tools: ledgersync_list_check_images and ledgersync_get_check_images_download_link. The images never enter the chat. The links work once and expire after 10 minutes, like statement links.
  • Name the files your way: say how, for example payee, then the date, then the check number. Up to 200 checks a link, and up to 10 links at a time.
  • Your AI app loads the new tools with the tool list (see Tool updates on For developers).
AI connector

Your AI explains what changed this month, and scopes a new client’s cleanup

Ask your AI “Why did Acme’s expenses jump in October?”. It compares two periods of the client’s QuickBooks Online profit and loss or balance sheet, names the vendors and customers behind the biggest changes, and writes a short summary you can send. See What changed this month.Ask “Check Acme’s books before I price the cleanup.”. After the figures, your AI asks for your firm’s rates and turns them into a scope with hours. See Check a new client’s books.
  • The steps are also MCP prompts, flux_and_summary and prospect_diagnostic. The client snapshot now recommends flux_and_summary while QuickBooks works, and the descriptions of ledgersync_get_financials, ledgersync_drill_account and ledgersync_run_diagnostic name their playbook.
  • No new tools. Your AI app loads the new prompts and descriptions with the tool list (see Tool updates on For developers).
AI connector

LedgerSync is in ChatGPT’s plugin directory

In ChatGPT, open Plugins, search LedgerSync and click Install plugin. Then sign in with your LedgerSync login and click Allow. No address to copy, and no Developer mode. See Set up in ChatGPT.
  • The plugin from the directory gets new LedgerSync tools on its own, after OpenAI’s daily check, so it has no Refresh tools to click (see After a LedgerSync update).
  • Added LedgerSync by hand before? It keeps working. To switch, see Added LedgerSync to ChatGPT by hand before? on Set up in ChatGPT.
  • The sandbox address is still added by hand (see For developers). When a tool list is out of date, the connector’s next_step now says Refresh tools only for a copy added by hand.
AI connector

Your AI checks a new client’s books before you price the cleanup

Ask your AI “Check Acme’s books before I price the cleanup.”. It reads the client’s QuickBooks Online books over the last 12 full months and gives the figures a cleanup is priced from: entries per month, each bank and card account with how long since an entry was reconciled, the entries waiting in Uncategorized and Ask My Accountant, suspense balances, unused accounts, and receivables and bills over 90 days. It gives figures, never hours: for an estimate, your AI asks for your firm’s own rates.
  • See ledgersync_run_diagnostic in For developers. Each account it lists carries the account_id that ledgersync_drill_account takes.
  • Your AI app loads the new tool list first (see Tool updates on For developers).
AI connector

Your AI asks clients about their transactions, in your words

Ask your AI “Email Acme and ask what these three charges were.”. It shows you the message and the transactions first. After your yes, LedgerSync sends it from its own address with a link to the same page as the request you send from the LedgerSync app: the client sees only those transactions and writes what each one was, and the answers are saved on the transactions in LedgerSync.
  • See ledgersync_preview_client_message with purpose: "clarification" in For developers. Your firm’s Clarification templates work too.
  • Later, ask “What did Acme answer?”. Your AI reads the answers back, and which transactions are still waiting, with ledgersync_list_clarification_answers.
  • Your AI app loads the new tool list first (see Tool updates on For developers).
AI connector

Your AI explains a line of a financial report, payee by payee

Ask your AI “Which vendors are behind the jump in Acme’s equipment rental in October?”. It breaks the account down by the vendors, customers and others on its entries, straight from the client’s QuickBooks Online: each one’s amount in each period, which are new and which stopped. It starts from a line of a financial report, with the report’s own dates, so the total matches the line.
  • See ledgersync_drill_account in For developers. The account lines of ledgersync_get_financials now carry the account_id it takes.
  • Your AI app loads the new tool list first (see Tool updates on For developers).
AI connector

Applying the same rules setup again says where it stands

When your AI applies a rules setup it already applied, LedgerSync no longer repeats its first answer (“queued”). It says the setup was already sent and gives its status now, such as done or undone. As before, nothing is applied twice.
  • The same goes for linking a global category at every client, and for marking vendors for 1099 (ledgersync_apply_ledger_changes).
  • See ledgersync_apply_rules_setup in For developers. No new tool: your AI app keeps the tool list it has.
AI connector

The reconciliation check is clearer when a statement agrees with the books

A statement that agrees with the client’s QuickBooks books no longer comes back with entries to check:
  • Rows of one date and one amount that nothing tells apart, such as several insurance payments of the same amount on one day, are now paired. They used to be listed as “not matched: check that they are one entry”.
  • LedgerSync also reads the statement before the previous one. An entry the books date a day after the bank posted it, across a month end, is no longer named as still open two months later.
  • Entries still open from before the statement that cancel each other out are named as such, instead of sending you to the previous statement.
  • See ledgersync_compare_statement_to_books in For developers. No new tool: your AI app keeps the tool list it has.
AI connector

Your AI pulls a client’s financial reports from QuickBooks

Ask your AI “Show Acme’s P&L for October, compared with September.”. It reads the report straight from the client’s QuickBooks Online, with QuickBooks’ own amounts: profit and loss, balance sheet, cash flow, trial balance, A/R aging or A/P aging. Compared with another period, each line shows its change, and the big changes are flagged with what moved. Your AI can then turn the report into a PDF or a spreadsheet to send. In an AI app that runs scheduled tasks, you can ask for it every month.
AI connector

Your AI walks you through why an account won’t reconcile

Ask your AI “Why won’t Acme’s Chase checking reconcile for September?”. It now follows the same steps every time: it finds the month’s statement, compares it with the client’s QuickBooks account, asks which QuickBooks account to use when the account is not linked, tells you the difference and its likely causes in plain words, lists what to fix in QuickBooks, and says what the check does not do. See Explain my reconciliation.
  • The steps are also an MCP prompt, explain_rec_difference. The client snapshot now recommends it, and the short menu a new connection shows includes it. See For developers.
  • The check now offers the previous statement only when part of the difference started before this statement and no entry explains it.
  • No new tools. The recommendation and the offers arrive in LedgerSync’s answers right away; the full steps reach your AI app once it reloads LedgerSync’s tools (see After a LedgerSync update).
AI connector

The reconciliation check names the likely causes

When your AI compares a bank or card statement with the client’s QuickBooks Online books, it now also names what made the difference, the strongest first, each with its evidence and its share of the difference: a bank fee or interest the books do not have, two swapped digits, an entry entered twice, a wrong sign, an entry dated on the other side of the statement’s end, or items the bank has not shown yet. Timing, such as a check not cashed yet, is told apart from mistakes.
  • See likely_causes under ledgersync_compare_statement_to_books in For developers.
  • No new tool: your AI app keeps the tool list it has.
AI connector

Your AI checks why a bank or card statement does not reconcile

Ask your AI “Why won’t Acme’s checking account reconcile for September?”. It compares LedgerSync’s copy of the statement with the client’s QuickBooks Online register for that account and month, says how far apart they are on the statement’s last day, and splits the difference into parts that always add up: what started before this statement, rows on the statement the books do not have (a bank fee, for example), entries in the books the bank has not shown yet (a check not cashed), and entries the books date after the statement ends. It shows the rows on each side that nothing matched. When the account is not linked to a QuickBooks account, it asks which one to use.
  • See ledgersync_compare_statement_to_books in For developers.
  • Your AI app loads the new tool list first (see Tool updates on For developers).
AI connector

Your AI reviews a client’s month in one go

Ask your AI “Review Acme’s books for October”. It now walks through all five checks on the month, one after another: entries still waiting for a category, expenses with no payee, entries on a parent account, entries coded differently than the vendor’s usual account, and amounts over $1,000. For each entry it shows the reason and, when the client’s history shows one, a suggested fix. It asks you about the ones only you can settle, lists the fixes for you to make in QuickBooks, and says what the review does not check. See Review a client’s books.
  • The steps are also an MCP prompt, month_end_review. The client snapshot now recommends and offers the review, and the short menu a new connection shows includes it. See For developers.
  • No new tools: your AI gets the new steps in LedgerSync’s answers, without loading a new tool list.
AI connector

Your AI reads your clients again before it loads LedgerSync’s new tools

Since October 30, LedgerSync asks your AI for each client’s name with the client’s id. An AI app that still had the older copy of LedgerSync’s tools could not send the name, so it could not look at any client. ChatGPT keeps that older copy until you click Refresh tools. Now it can still read a client’s accounts, statements, transactions, connections, rules and QuickBooks details without the name, and each answer says which client it is about. Previews and changes still ask you to load the new tools first.
  • The one-client reads that existed before October 30 answer without client_name. Previews, the categorized export link, refreshes, fix and connect links, changes and tools added since still need it. See For developers.
  • To use everything again, load the new tool list once: see After a LedgerSync update.
AI connector

Your AI updates a client’s phone numbers, emails and name

Ask your AI “Add 431 432 4232 as Will’s phone number”. It now changes a client’s details in LedgerSync, as the Edit Client screen does in the app: the client’s name, main email, main phone number (the one LedgerSync texts) and extra phone numbers, and it can add extra emails. LedgerSync tells your AI to say exactly what will change and to ask for your yes first. A client added through your AI can now get its phone number this way, so LedgerSync can text it an invitation to connect its bank.Changing the email a client signs in to LedgerSync with can also move its sign-in, so your AI asks you first. Once it moves, the client signs in with the new email, LedgerSync emails a sign-in link to the new address and a notice to the old one, and the client’s old password, two-step sign-in and phone app sign-ins stop working.
  • See ledgersync_get_client and ledgersync_update_client in For developers.
  • Your AI app loads the new tool list first (see Tool updates on For developers).
AI connector

Your AI spots entries coded differently than usual, and shows how a vendor is usually coded

Ask your AI “Review Acme’s books for October”. It now runs a fifth check: entries booked to an account the vendor rarely uses, such as a Home Depot purchase in Meals when Home Depot’s entries almost always go to Job Supplies, with the usual account as the suggested fix. Ask “How is Home Depot usually coded for Acme?” and it shows the accounts the vendor’s entries went to over the last 12 months, most used first, with how often, when last and the average amount. It can also start from the text of a bank row.
  • See expense_inconsistency and ledgersync_get_vendor_coding_history in For developers.
  • The client snapshot now also says how many entries each check flagged, when a check read the client’s books for last month today.
  • An expense with no payee now also gets its suggested payee from the bank rows LedgerSync sent to QuickBooks with the same text.
  • Your AI app loads the new tool list first (see Tool updates on For developers).
v2 API

The v2 API reference and guides are now on this site

The new v2 API (legacy) tab, next to the v3 docs, holds the v2 API: five guides (overview, authentication, transactions, transaction status and testing) and an endpoint reference built from the live v2 spec. The reference is checked against the live spec every 6 hours, so it follows the API as it changes.
  • The v2 sandbox is https://staging.ledgersyncappv2.com, with the same paths as production, https://ledgersyncappv2.com. The old sandbox address, 104.196.54.242.sslip.io, no longer works.
  • Start with the v2 overview.
AI connector

Your AI checks a client’s books for common mistakes

Ask your AI “Review Acme’s books for October”. It now checks four things in your client’s QuickBooks Online: entries still waiting in Uncategorized or Ask My Accountant, expenses with no payee, entries booked to a parent account instead of one of its sub-accounts, and large amounts. Each one comes with why it was flagged and, where the client’s history or the entry’s own text makes it clear, a suggested fix that you then make in QuickBooks.
AI connector

Your AI starts work on a client with one look at where things stand

Ask your AI “Where do things stand with Acme?”. In one step it now sees which accounting software the client uses, whether its QuickBooks connection works, the date its books are closed through in QuickBooks, its bank and card accounts with each one’s latest statement, and what it can help with next, such as finding who needs a 1099 in the season.
AI connector

Every firm can now mark vendors for 1099 from their AI

After your yes, your AI marks the vendors who need a 1099 in your client’s QuickBooks, and can put them back. This was open to a few test firms; it is now open to every firm. No reconnect needed. See Find who needs a 1099.
AI connector

Your AI advises which vendors need a 1099, from the expense accounts they were paid through

Ask your AI “Which vendors need a 1099 for Acme?”. Next to each vendor, it now says whether the expense accounts its payments were booked to, such as Contract Labor, Legal Fees, Rent or Supplies, point to a 1099, point away, or need your answer, with the reason. It asks you about the unclear ones first. You can also tell it to count only some accounts, the way QuickBooks’ own 1099 setup does. You decide. See Find who needs a 1099.
  • See advice, advice_reason, expense_accounts and count_only_account_ids in For developers. The default page of the 1099 list is now 15 vendors.
  • The advice reaches your AI with no reconnect. To count only some accounts, and to get the updated form_1099_season steps, your AI app loads the new tool list first (see Tool updates on For developers).
AI connector

Your AI names each client, and LedgerSync stops it from using the wrong one

Your AI now gives LedgerSync each client’s name with the client’s id. LedgerSync checks they are the same client before it reads or changes anything. If the AI uses another client’s id, for example one it remembered from an earlier chat, LedgerSync refuses it, and the AI looks the client up again. Every answer names the client it is about, so you always see whose data you are looking at.The same goes for a client’s categories, payees and vendors: one taken from another client’s list is refused, and the AI gets this client’s list again.
  • Every tool that works on a client takes client_name with client_id (client_names with client_ids), and ledgersync_get_fix_link now takes client_id. See For developers.
  • Your AI app loads the new tool list once: connect again in Claude, click Refresh tools in ChatGPT, or Reauthenticate in Grok, then start a new chat. See After a LedgerSync update.
AI connector

Your AI accepts For review transactions, and checks your rules

Ask your AI to move a client’s For review transactions in the Rules Engine to Categorized, as Accept does in the LedgerSync app. For example: “For Acme, Beta and Cedar, move the transactions with rules to Categorized, except exclusion rules and the ones coded to 8001 or 8002. Give me a report of what you did and didn’t move.” It shows you what would move and what would stay first, moves them only after your yes, and then tells you what moved and what did not, and why. It can undo it too. For Manual and QuickBooks Online clients, up to 10 clients at a time. Nothing is posted to QuickBooks. See Accept For review transactions.Your AI can also check your rules against the transactions they filed, and tell you about every mismatch. And after it sets up rules, it offers to accept the transactions they filled.
  • See ledgersync_list_review_transactions, ledgersync_preview_accept_transactions, ledgersync_accept_transactions and ledgersync_undo_accept_transactions in For developers.
  • To use the new tools, your AI app loads the new tool list first (see Tool updates on For developers).
AI connector

Your AI runs a client’s 1099 check the same way every time

Ask your AI “Which vendors need a 1099 for Acme?”. It lists the vendors who need a 1099, goes through the ones not marked for 1099 in QuickBooks and the ones with no tax ID on file, offers to mark them after your yes (marking opens to a few test firms first), and ends with what the list does not cover. The menu your AI shows on a new connection can now offer this question too. See Find who needs a 1099.
  • The steps are also an MCP prompt, form_1099_season. See For developers.
  • The list’s new ending and the menu reach your AI with no reconnect. The prompt shows after your AI app connects again.
AI connector

Your AI lists a client’s 1099 vendors, and can mark them in QuickBooks

Ask your AI who needs a 1099 for a QuickBooks Online client: it lists the vendors paid at or over the 1099-NEC threshold in the tax year (2,000.00 for 2026), with whether each one is marked for 1099 in QuickBooks and whether a tax ID is on file. Card and payment app payments are left out, since they go on a 1099-K. A tax ID is never shown.Your AI can also mark those vendors for 1099 in QuickBooks: it shows each vendor’s 1099 flag before and after, marks them only after your yes, and can put them back. Marking opens to a few test firms first.
  • See ledgersync_get_1099_summary, ledgersync_preview_ledger_changes, ledgersync_apply_ledger_changes, ledgersync_get_ledger_changes_status and ledgersync_undo_ledger_changes in For developers.
  • To use the new tools, your AI app loads the new tool list first (see Tool updates on For developers).
AI connector

Rule changes and undo keep hidden accounts out too

When your AI previews a change to an existing rule or an undo, and when it reports how a rules setup went, its counts and example transactions now leave out the accounts hidden in LedgerSync, like the rest of its rule suggestions. Transactions on a hidden account that a changed rule had filed are filed again by your other rules, the same way as on the accounts you see.
  • See ledgersync_preview_rules_setup, ledgersync_get_rules_setup_status and ledgersync_undo_rules_setup in For developers.
  • Your AI keeps working as before, with no reconnect.
AI connector

Excluded transactions stay out of your AI’s transaction lists

When your AI lists an account’s transactions, it now leaves out the ones on LedgerSync’s Excluded list, unless you ask for excluded transactions. Then it lists them too, and says which ones are excluded.
  • ledgersync_list_transactions takes include_excluded, and marks each excluded transaction excluded. See For developers.
  • Excluded transactions stay out with no reconnect. To ask for them, your AI app loads the new tool list first (see Tool updates on For developers).
AI connector

Hidden accounts stay out of your AI’s rule suggestions

Your AI’s list of a client’s transactions that no rule has filed yet now leaves out the accounts hidden in LedgerSync, the way the LedgerSync app does. The count of those transactions, and what a new rule would file, leave them out too, so your AI no longer brings up transactions from accounts you cannot see for that client.
  • ledgersync_list_uncategorized_transactions leaves out transactions on hidden accounts. See For developers.
  • Your AI keeps working as before, with no reconnect.
AI connector
When your AI has nothing else to go on for a QuickBooks client’s category link, LedgerSync now looks at how the client’s QuickBooks booked its expenses and deposits over the last 12 months. See Link your global categories.
  • Proposals instead of questions. If your global rules for the category match at least 3 of those transactions, and at least 8 in 10 were booked to one of the client’s categories, the AI proposes that one.
  • Best guesses stay first. When QuickBooks points elsewhere than a best guess from the names or your other clients’ links, the AI shows it as another option.
  • One client at a time. LedgerSync reads a client’s QuickBooks only when the AI needs it, and keeps that read for an hour. Transactions LedgerSync sent to QuickBooks itself do not count.
  • ledgersync_list_setup_categories has booked_in_quickbooks on candidates and quickbooks_history on the page. See For developers.
  • Your AI keeps working as before, with no reconnect.
AI connector
When your AI links a QuickBooks client’s categories to your global categories, it now asks you only about the links that the client’s waiting transactions need. See Link your global categories.
  • The links that matter first. The global categories the client’s waiting transactions are waiting for come first, with how many transactions wait for each.
  • Fewer questions. When the AI has nothing to go on for a link that no waiting transaction needs yet, it no longer asks. It tells you how many can wait: link them later, or ask for them now.
  • ledgersync_list_setup_categories (kind firm) has a new field, waiting_on_link. See For developers.
  • Your AI keeps working as before, with no reconnect.
AI connector

New QuickBooks categories under another category, with a detail type

Your AI can now create a QuickBooks client’s new category under one of the client’s categories, and give it a QuickBooks detail type. It shows you a preview first and saves nothing until you say yes. See QuickBooks clients.
  • Under a category you have. For example: Make a new category Fuel under Auto. QuickBooks calls it a sub-account. The preview, the result and the undo show it with that category’s name in front: Auto:Fuel.
  • Checked in QuickBooks first. The preview checks the category it goes under in the client’s QuickBooks: it must be active, and QuickBooks allows 5 levels at most. The new category takes its type, so it needs no type of its own.
  • Detail types. A new category can have a QuickBooks detail type, such as Travel or Office/General Administrative Expenses, spelled as QuickBooks shows it. Under another category, a new one without a detail type gets that category’s. At the top level the AI gives every new category one, since QuickBooks would otherwise pick its own (Travel, for an expense category). The preview shows the detail type each new category gets.
  • The same name in another place. A category with the same name at the top level does not stop a new one under another category, and two new categories can share a name under different ones.
  • QuickBooks clients only. A Manual client’s new categories cannot go under another one or have a detail type.
  • Not yet: a new category under another new one. Set up the first one, then ask for the one under it.
  • New input on ledgersync_preview_rules_setup: parent and detail_type on a new client category. New refusal codes: parent_invalid and detail_type_invalid. See For developers.
  • Your AI keeps working as before. To create a category under another one, or with a detail type, it needs the new tool list: reconnect in Claude, click Refresh tools in ChatGPT, or reauthenticate in Grok, then start a new chat. See After a LedgerSync update.
AI connector

A greeting with your firm’s name, and which AI plans work

  • A greeting. When your AI checks your LedgerSync account, it greets you with your firm’s name and says whether this is live or sandbox, and whether it can make changes or only read. The first time, it also shows a few things you can ask, like which bank connections need attention.
  • Which plans work. Claude: any plan, and on the free plan LedgerSync is your one custom connector. ChatGPT: any plan, even the free one, on chatgpt.com in a web browser. See Start here.
  • ChatGPT’s newer screens. You add LedgerSync with Add, then Add custom MCP server, and load new tools with Refresh tools (Settings, then Plugins, then LedgerSync). See Set up in ChatGPT and After a LedgerSync update.
  • Nothing to reload: the greeting comes with the account check your AI already runs (ledgersync_get_account_status, which now returns next_step and, the first time, suggested_questions).
AI connector
Your AI can now link one of your firm’s global categories at every client that has no category linked to it yet, in one preview. It proposes the client’s category for each one, and saves nothing until you say yes. See Link your global categories.
  • One preview for all your clients. The AI shows the totals and a summary of the categories it would link, for example Fuel at 120 clients, Auto:Fuel at 40, then the clients that need your answer.
  • Change only what is wrong. Say yes, or name the clients to change or leave out. The AI checks again with your changes.
  • QuickBooks clients. Once a client’s category is linked, your global rules file the transactions that were waiting for it, and the “map this category” badge goes away. The preview says how many.
  • Manual clients too. There, a link makes LedgerSync use the GL code of the client’s own category.
  • Your clients only. An employee links the category only at the clients assigned to them.
  • One undo for all. One undo takes back every link the setup made, unless someone changed it since. At QuickBooks clients, the transactions filed through those links go back to waiting.
  • New tool: ledgersync_preview_category_links. ledgersync_apply_rules_setup saves its preview, and ledgersync_get_rules_setup_status and ledgersync_undo_rules_setup take its setup_id (rsl_...) without a client_id. See For developers.
  • Your AI keeps working as before. To link a category at every client, it needs the new tool list: reconnect in Claude, click Refresh in ChatGPT, or reauthenticate in Grok, then start a new chat. On ChatGPT Business, Enterprise or Edu, your workspace admin refreshes LedgerSync’s tools, and on Enterprise and Edu also turns on the new tool under Action control. See After a LedgerSync update.
Accounts

What is still owed on a card’s last statement

statement_balance is the balance printed on the last statement, so it does not go down when the card is paid. Two new fields on a credit card’s liabilities say how much of it is still unpaid:
  • statement_date: the day that statement closed, the same date as its statement_date in GET /v3/accounts/{account_id}/statements.
  • remaining_statement_balance: statement_balance minus every payment, refund and other credit posted after statement_date. Negative like statement_balance, 0 once the statement is paid off or when it closed in credit, and never more than the card owes in total. Pending payments count once they post.
No bank or aggregator reports this figure, so LedgerSync computes it, for credit cards from Finicity and MX. The two fields are null together for a few days after each close, until the new statement reaches us, and on cards with no statement on file. They are always null in sandbox. See What is still owed on the last statement.
AI connector

Change your global rules too

Your AI can now edit, move and delete your firm’s global rules, not only turn them off or on. It shows you a preview first and saves nothing until you say yes. See Changing your rules.
  • Every client at once. A change to a global rule works at every client of your firm with the same accounting app. The preview shows what happens at your other clients.
  • Transactions follow the change. The transactions a global rule no longer files go back to each client’s other rules right away. Transactions someone changed by hand stay as they are.
  • QuickBooks clients. Turning off, deleting or editing a global rule takes off its “map this category” badges. A global rule you edit or turn on files transactions only at the clients that link its global category. At the others, its transactions wait with the badge, and the preview names those clients.
  • Rules that share a priority. When some of your global rules have the same Priority number, the AI cannot move a global rule across or next to them, or delete one of them. Ask it for another place, or move or delete the rule in the LedgerSync app.
  • One-step undo. Undoing the setup puts a global rule back as it was, a deleted one included, and its transactions at every client. A deleted rule’s earlier lines on the Apply Status tab do not come back.
  • Changing global rules needs Manage Global Rules.
  • New result fields, for a QuickBooks global rule a change edits or turns on: waiting_on_link_here, waiting_on_link_other_clients and clients_needing_link_count in the preview’s changes, with its clients in clients_needing_link, and waiting_on_link_here, waiting_on_link_other_clients and other_clients_needing_link in the status’s result.changes. No new tool. See For developers.
  • If your AI still says a global rule can only be turned off or on, it has the old tool descriptions: reconnect in Claude, click Refresh in ChatGPT, or reauthenticate in Grok, then start a new chat. See After a LedgerSync update.
AI connector

Change a client’s rules from Claude, ChatGPT and Grok

Your AI can now change the rules you already have, not only add new ones. Ask it to edit a rule, turn it off or on, move it or delete it. It shows you a preview first and saves nothing until you say yes. See Changing your rules.
  • Client rules: every change. Edit a rule’s name, money in or out, conditions, category or vendor, turn it off or on, move it in the list, or delete it. For Manual and QuickBooks Online clients.
  • Global rules: off or on. For now, the AI can turn a global rule off or on. Edit, move or delete global rules in the LedgerSync app.
  • Exclusion rules too. You can change an exclusion rule the same way. It stays an exclusion rule: the waiting transactions it keeps or fills are marked for exclusion, and move to Excluded when you accept them.
  • Transactions follow the change. When you turn a rule off, delete it, or edit it so it no longer matches some of its waiting transactions, the client’s other rules file those again right away, as they do new transactions. Transactions someone changed by hand stay as they are. The preview shows the rule before and after, and what happens to its transactions.
  • Rules on the amount. A rule can now also look at the amount: equals, doesn’t equal, is greater than or is less than. For example: a rule for Shell over 500. In new rules and in changes. A rule still needs words to look for in the bank text.
  • One-step undo. Undoing the setup puts the rules back as they were, a deleted rule included, and their waiting transactions too, unless someone changed them since.
  • History, not LS-AI. A changed rule is not marked LS-AI. Its Rule History in the LedgerSync app shows the change (a deleted rule has no page to open its history from).
  • Not yet: editing, moving or deleting global rules, changing the category of a rule split over several categories, and taking a rule’s vendor off.
  • New tool: ledgersync_get_rule, which shows one rule in full. New input on ledgersync_preview_rules_setup: changes and the amount operators. New result fields: the changes in the preview, the status and the undo check. See For developers.
  • Your AI keeps working as before. To change rules, it needs the new tool list: reconnect in Claude, click Refresh in ChatGPT, or reauthenticate in Grok, then start a new chat. See After a LedgerSync update.
AI connector

The AI offers rules on its own

You no longer need to ask your AI for rules, or for a global rule. When you ask it to show a client’s bank transactions and some of the client’s transactions have no rule yet, it offers to suggest rules for them. Say yes, and it sets them up with you. See Set up rules for a client.
  • It names each rule. The AI proposes each rule by name and says what it does and for whom, for example: Can I create a global rule ‘Square’ that leaves Square deposits out of the books for every client? You say yes, no, or what to change.
  • It picks client or global. A global rule for vendors that mean the same for every client, such as bank fees, software or fuel, and a client rule otherwise. A global rule still needs Manage Global Rules.
  • Exclusion rules too. When transactions look like they should be left out of the books, such as transfers between the client’s own accounts, it proposes an exclusion rule itself.
  • After a QuickBooks push. In Bookkeeping into QuickBooks, when some of the transactions had no rule, the AI offers rules so transactions like them file themselves next time.
  • Nothing changes before your yes. As before, the AI shows you a preview first and saves nothing until you say yes.
  • New result field: transactions_without_rule on ledgersync_list_transactions, for Manual and QuickBooks Online clients. See For developers.
  • Your AI keeps working without reconnecting. To get the updated tool descriptions, reconnect in Claude, click Refresh in ChatGPT, or reauthenticate in Grok, then start a new chat. See After a LedgerSync update.
AI connector

Export categorized transactions on a schedule from Claude

For Manual clients. Your AI gives you a link to a file of the transactions you accepted in LedgerSync that were not exported yet: the same file as Export Transactions in the LedgerSync app, Default or ACS. Make it a scheduled task in Claude, for example every weekday at 3 PM. In ChatGPT, you may need to ask in a chat: its scheduled tasks may not reach LedgerSync yet. LedgerSync marks the transactions Synced when you click Download, so the next file has only the new ones. See Export categorized transactions on a schedule, which has the instructions to paste into your AI.One new AI connector tool, on the live and sandbox connectors:
  • ledgersync_get_categorized_export_link says how many categorized transactions are waiting, for every Manual client you manage or for one client, and gives a link to them as the app’s Default (CSV) or ACS (Excel) file. Calling it changes nothing.
Limits and safeguards:
  • Nothing is marked Synced until a person clicks Download on the link’s page. A link nobody opens changes nothing, and the next link has those transactions too.
  • A link works for 24 hours. Clicking Download again gives the same file. Each click checks the connection again. Anyone with the link can download the file, so don’t forward it.
  • One file holds up to 10,000 transactions. Each connected AI app can make 30 links an hour, and each link takes 5 clicks a minute.
  • Clients connected to QuickBooks, Zoho Books or another accounting app are left out.
  • Claude runs the schedule, on paid plans: LedgerSync stores none.
  • To see the new tool, reconnect in Claude, click Refresh in ChatGPT, or reauthenticate in Grok, then start a new chat. See After a LedgerSync update.
See For developers.
AI connector

The AI can set up exclusion rules

Rules setup can now make exclusion rules too, for the transactions you want left out of the books, such as transfers between the client’s own accounts. The AI asks you first, and they go in the same preview as your other rules. See Set up rules for a client.
  • Like the exclusion rules you make in the app. An exclusion rule looks for words in the bank text and has a category, but needs no vendor. It can be a client rule or a global rule, for Manual and QuickBooks Online clients.
  • Marked, then excluded when you accept. The waiting transactions an exclusion rule fills stay on For review, marked for exclusion. When you accept them, they move to Excluded instead of Categorized.
  • The preview and the result count them. The preview says how many waiting transactions each exclusion rule would mark, for this client and at your other clients, and the result says how many it marked.
  • Undo puts them back. Transactions you moved to Excluded because of the setup’s exclusion rules go back to For review. The AI says how many before you say yes.
  • New field: rule_type (inclusion, the default, or exclusion) on ledgersync_preview_rules_setup. New result fields: the rule type and the exclusion counts in the preview, the status and the undo check. See For developers.
  • Your AI keeps working without reconnecting. To get the updated tool descriptions, reconnect in Claude, click Refresh in ChatGPT, or reauthenticate in Grok, then start a new chat. See After a LedgerSync update.
AI connector

Set up a QuickBooks client’s rules from Claude, ChatGPT and Grok

Rules setup now works for QuickBooks Online clients too, with the same nine tools. The AI proposes rules for the client’s waiting transactions and, after you say yes, LedgerSync saves them and fills the transactions. See Set up rules for a client.
  • All the missing links in one preview. For each global category the client has no category linked to, the AI proposes a link: from the client’s own history (where its past transactions were booked), from a best guess (the same name, or the category your other clients link it to), or it asks you when there is nothing to go on. This works for Manual clients too.
  • Why links matter at a QuickBooks client. A global rule files a QuickBooks client’s transaction only through the client’s category linked to the global category. Without a link, the transaction waits with the “map this category” badge. The preview says how many transactions each link lets your global rules fill, and names your other QuickBooks clients that still need a link: only the ones you can see in LedgerSync.
  • Created in QuickBooks after your yes. The client’s new categories and vendors are created in its QuickBooks before the rules are saved. A name QuickBooks already has is used instead, and a deactivated category or vendor is never turned back on.
  • Customers as payees. A client rule can pay one of the client’s QuickBooks customers. Not employees.
  • Undo. It removes the setup’s rules and links. The transactions your global rules filed through those links go back to waiting. Categories and vendors created in QuickBooks stay there.
  • A broken QuickBooks connection. Links and rules still work. New categories, new vendors and customers as payees wait until you reconnect QuickBooks.
  • Then post them into QuickBooks. When the setup is done, the AI offers to post the filled transactions with Bookkeeping into QuickBooks.
  • No new tools. New result fields: the link suggestions on ledgersync_list_setup_categories, the customers on ledgersync_list_setup_vendors, and the waiting-for-a-link counts and other clients in the preview and the status. See For developers.
  • Your AI keeps working without reconnecting. To get the updated tool descriptions, reconnect in Claude, click Refresh in ChatGPT, or reauthenticate in Grok, then start a new chat. See After a LedgerSync update.
AI connector

Set up a client’s rules from Claude, ChatGPT and Grok

For Manual clients. The AI reads the client’s categories, vendors and rules, your firm’s global categories, global vendors and global rules, and the client’s waiting transactions that no rule fills yet. It proposes rules and asks you, for each one, whether it is for this client only or for every client of your firm (a global rule). It shows you a preview first. After you say yes, LedgerSync saves the rules, with any new categories and vendors, and fills the waiting transactions. Everything the AI makes is marked LS-AI and can be undone. See Set up rules for a client, which has the instructions to paste into your AI.Nine new AI connector tools, on the live and sandbox connectors:
  • ledgersync_get_rules_setup checks whether the client’s rules can be set up, and counts what it already has.
  • ledgersync_list_setup_categories, ledgersync_list_setup_vendors and ledgersync_list_rules read the client’s and the firm’s categories, vendors and rules.
  • ledgersync_list_uncategorized_transactions groups the client’s waiting transactions that no rule fills by their bank text.
  • ledgersync_preview_rules_setup checks the proposed rules and counts what each one would fill, for this client and at your other clients. It changes nothing.
  • ledgersync_apply_rules_setup saves a previewed setup, only after you say yes.
  • ledgersync_get_rules_setup_status says what was created and filled, and what failed and why.
  • ledgersync_undo_rules_setup shows what an undo would remove and keep, then undoes the setup after you say yes.
Limits and safeguards:
  • Nothing is saved without a preview. A preview’s confirmation works for 30 minutes, and only for the person and the AI app that made it.
  • Your existing rules stay as they are. A setup never changes them, and fills only transactions that no rule filled and nobody gave a category.
  • One preview takes up to 100 rules, 50 new global categories, 50 new client categories, 100 category links and 100 new vendors. Each connected AI app can make 60 previews an hour, 30 setups a day and 10 undos a day.
  • Global rules, global categories and global vendors need the Manage Global Rules permission.
  • Not yet: QuickBooks clients (they come next), changing or turning off existing rules, amount conditions, and clients on Zoho Books.
  • In Claude Code, the instructions are also a command: rules_setup.
  • To see the new tools, reconnect in Claude, click Refresh in ChatGPT, or reauthenticate in Grok, then start a new chat. See After a LedgerSync update.
See For developers.
StatementsAI connector

Converted statements say why their transactions don’t add up

A converted statement used to say only whether its transactions match the statement’s balances. It now says why when they don’t.
  • POST /v3/statements/extract: the operation result has a new validation_status: balanced, balance_mismatch, totals_mismatch, balance_missing or no_transactions. validation_message says the same in a short plain message for the person who uploaded the statement, and validated is true only for balanced. Extractions completed before this change have no validation_status.
  • AI connector: ledgersync_get_statement_conversion returns validation_status too, and asks the AI to repeat validation_message when the transactions don’t add up.
  • Loan statements now add up when the balance moves by the principal of each payment, and an extra account with no transactions and a missing balance no longer stops a statement from adding up.
AI connector

Bookkeeping into QuickBooks from Claude, ChatGPT and Grok

For clients whose bank transactions do not come into QuickBooks through a bank feed. The AI picks a category and a payee for each of the client’s transactions waiting in LedgerSync, asks you only about the ones it is not sure of, and shows you a preview. After you say yes, LedgerSync posts them into the client’s QuickBooks, with the same upload as a sync from the LedgerSync app. See Bookkeeping into QuickBooks, which has the instructions to paste into your AI.Seven new AI connector tools, on the live and sandbox connectors:
  • ledgersync_get_quickbooks_setup checks the client’s QuickBooks connection and which QuickBooks bank account each account posts to.
  • ledgersync_list_transactions_to_categorize lists an account’s transactions waiting on the For review list, with LedgerSync’s suggestions.
  • ledgersync_list_quickbooks_categories and ledgersync_search_quickbooks_payees read the client’s QuickBooks categories and payees.
  • ledgersync_preview_quickbooks_push checks everything and shows what would be posted. It changes nothing.
  • ledgersync_push_to_quickbooks posts a previewed set, only after you say yes.
  • ledgersync_get_quickbooks_push_status says what was posted, skipped or failed, and why.
Limits and safeguards:
  • Nothing is posted without a preview. A preview’s confirmation works for 30 minutes, and only for the person and the AI app that made it.
  • The AI keeps what your rules already chose unless you change it. When one of your global rules matched a transaction but the client has no QuickBooks category mapped to the rule’s global category, the AI asks you which category to use.
  • One push takes up to 100 transactions, 10 new payees (added to QuickBooks as vendors) and 5 new account mappings. Each connected AI app can make 120 previews an hour and 60 pushes a day.
  • Not yet: transfers, splits, classes, bill or invoice payments, and accounts on LedgerSync’s own bank connection.
  • On the sandbox connector, a push really posts, to an Intuit sandbox QuickBooks company.
  • In Claude Code, the instructions are also a command: quickbooks_bookkeeping.
  • To see the new tools, reconnect in Claude, click Refresh in ChatGPT, or reauthenticate in Grok, then start a new chat. See After a LedgerSync update.
See For developers.
ChecksFDE

Correct a check’s OCR readings

PATCH /v3/accounts/{account_id}/checks/{check_id} corrects what LedgerSync read off a check image: the payee, amount, check number, check date, bank name and memo in the check’s ocr object. The correction is saved in LedgerSync, the same as when a firm fixes the check in the app.
  • Send only the readings you want to change, inside ocr. Anything you leave out keeps its value, and null clears a reading.
  • The bank-reported fields stay read-only.
  • A correction is kept when the bank feed refreshes the check later, whether it was made through the API or in the app.
  • Requires the new write:checks scope. Keys created in the developer portal have it from now on; add it to an existing key with PATCH /v3/api-keys/{id}.
See Correct a check.
AI connector

Use LedgerSync in Grok Bot

Grok Bot, Grok’s agent with its own computer, can now use LedgerSync. Before, it could add LedgerSync and sign in, but then it only got This AI app is not supported yet.
  • Ask your bot: “Add a custom remote MCP server named LedgerSync with URL https://api.ledgersyncappv2.com/mcp”. Then sign in with your normal LedgerSync login.
  • The sign-in asks you to allow Cursor. That’s normal: Grok Bot signs in through Cursor.
  • Grok Bot may change things in LedgerSync without asking you first, so tell it exactly what you want.
  • Already added it and got that message? Ask your bot again. It works now, without adding it again.
See Set up in Grok Bot.
AI connector
  • Links to a statement, a converted file and the upload page are now short. ChatGPT writes each link out again in its answer, and it sometimes changed a character of the old, long links, so they did not open.
  • A download link that no longer works opens a page that says what to do: go back to your chat and ask the AI for the file again. See Troubleshooting.
AI connector

Use LedgerSync in Grok

Add LedgerSync to Grok on grok.com and ask in plain words, as in Claude and ChatGPT. Sign in with your normal LedgerSync login: the firm owner, administrators and employees each see what they see in the LedgerSync app.
  • Works on any Grok plan, even the free one, on grok.com in a web browser: open grok.com/connectors, click New Connector, then Custom, and paste https://api.ledgersyncappv2.com/mcp.
  • On Grok Business and Enterprise, xAI asks a team admin to add LedgerSync in the xAI console first. Then each person connects it with their own LedgerSync login.
  • Grok may change things in LedgerSync without asking you first, so tell it exactly what you want.
  • To load a LedgerSync update in Grok, connect again: open grok.com/connectors, click LedgerSync, then Reauthenticate.
See Set up in Grok.
AI connector

Better QuickBooks files from converted statements, and deleted conversions

  • QuickBooks (.qbo) and OFX files of a converted statement now carry the statement’s ending balance when it prints running balances that add up, and give each transaction the same id in every file, so QuickBooks skips a transaction it already imported instead of adding it twice.
  • If you delete a conversion on your PDF Converter page, the AI now says it was deleted, instead of still showing its transactions and making a file link that does not work.
AI connector

Invite clients to connect their bank from Claude and ChatGPT

Ask the AI to invite a client to connect their bank, for example a client you just added. You write the email or text, or start from a template, and LedgerSync sends it with a link that signs the client in to their LedgerSync account.
  • ledgersync_preview_client_message takes two new settings. purpose is reconnect (the default, as before) or invite. client_ids names the clients to invite, up to 200. Any client you see can be invited, even one with banks already connected: the preview shows how many connections each one has.
  • Each invitation carries that client’s own sign-in link where the text says #LS_URL#. It signs the client in with nothing to type, and they connect their bank there. The link works for 6 days.
  • An invitation can use #ACCOUNTANT_NAME# (the full name of the person sending), but not #BANK_NAME#. An invitation email without a subject gets one like “Acme CPA invites you to connect your bank”.
  • An invitation can start from one of your firm’s Invitation templates in the Communication Center. ledgersync_list_message_templates now lists Issue and Invitation templates, and ledgersync_save_message_template takes purpose too.
  • A client already invited through the AI in the last 24 hours is skipped unless you ask to include them. Invitations count toward the same daily limit as other client messages: 300 a day per firm on live (100 on sandbox).
  • As with other client messages, nothing is sent until you confirm the preview, an employee invites only the clients assigned to them, and the sandbox connector never delivers.
  • Clients added through Claude or ChatGPT now work like clients added in the app: they can sign in to LedgerSync, and a bank they connect brings your firm’s usual number of past statements (before, some banks brought only one).
  • When you ask the AI to email or text a client, the connector now points it to LedgerSync’s preview and send, never to your own email app. LedgerSync sends the message with its own link for that client.
  • Claude and ChatGPT keep a copy of LedgerSync’s tools. To get the new settings, reconnect in Claude or click Refresh in ChatGPT, then start a new chat. See After a LedgerSync update.
See For developers.
AI connector

Turn a bank statement PDF into transactions from Claude and ChatGPT

Ask the AI to convert a bank statement PDF. It gives you a secure LedgerSync page to upload it, then shows the transactions in the chat and can give you a CSV, Excel, QuickBooks or OFX file. It is the same reader as the PDF Converter in the LedgerSync app, and the conversion also shows on your PDF Converter page there. See Turn a bank statement PDF into transactions.Three new AI connector tools, on the live and sandbox connectors:
  • ledgersync_create_statement_upload_link gives the upload page link. Each link takes one PDF (up to 30 MB) and works for 24 hours.
  • ledgersync_get_statement_conversion says when the conversion is done and returns the transactions, a page at a time, and whether they match the statement’s balances.
  • ledgersync_get_conversion_download_link gives a one-time, 10-minute link to the file.
Limits and safeguards:
  • Your firm can convert 3 statements at a time and 500 a day on live (2 and 50 on sandbox), shared with its API keys. Each AI app can make 30 upload links an hour.
  • Each person’s conversions are their own. A read-only connection cannot upload.
  • The AI cannot pass a file to LedgerSync, even one attached to the chat: the PDF goes through the upload page.
  • To see the new tools, reconnect in Claude or click Refresh in ChatGPT, then start a new chat. See After a LedgerSync update.
See For developers.
AI connector
A link from ledgersync_create_connect_link now shows the same bank list as the Add Bank Connection screen in the LedgerSync app, asks what to connect (transactions, statements, checks), and picks the connections the same way the app does.
  • With for_check_images: true, Checks comes already ticked. The link no longer lists only banks with check images, and it no longer leaves out the bank’s transactions: tick Transactions or Statements too and the page connects them in the same visit.
  • When the choice needs more than one connection, for example Regions Bank statements and checks, the page goes through them one at a time, and each sign-in opens in a new tab.
  • Links that developers make through the API keep the bank search they had.
  • Your AI keeps working without reconnecting. To get the updated tool descriptions, reconnect in Claude or click Refresh in ChatGPT, then start a new chat. See After a LedgerSync update.
AI connector

Email or text clients about broken bank connections from Claude and ChatGPT

Ask the AI to write to the clients whose bank connection needs attention, for example everyone with a broken American Express connection. LedgerSync sends your own email or text, and each message carries that client’s own link.Four new AI connector tools, on the live and sandbox connectors:
  • ledgersync_preview_client_message shows who will get the message, who is skipped and why, and a sample. Nothing is sent.
  • ledgersync_send_client_message sends the previewed message after you confirm. Each client gets a link to fix their connection or, for a bank LedgerSync now connects through MX such as American Express, a link to connect that bank again through MX.
  • ledgersync_list_message_templates and ledgersync_save_message_template read and add message templates. They are the same templates as in the Communication Center of the LedgerSync app, shared with the whole firm. An email template can now have a subject, which the AI uses when it sends.
Limits and safeguards:
  • Client messages are limited to 300 a day per firm on live (100 on sandbox).
  • A client who got a LedgerSync message about the same connection in the last 24 hours is skipped unless you ask to include them. Texts go out only in daytime hours: by default 11 am to 9 pm US Eastern time.
  • A message can carry no web address other than the client’s own LedgerSync link.
  • On the sandbox connector, client messages are never delivered and no link is created for them: the send shows what would be sent.
  • An employee messages only the clients assigned to them.
  • Claude and ChatGPT keep a copy of LedgerSync’s tools. To get the new tools, reconnect in Claude or click Refresh in ChatGPT, then start a new chat. See After a LedgerSync update.
See For developers.
AI connector

Connect a bank’s check images from Claude or ChatGPT

Ask the AI for a link to connect a client’s check images, for example: “Send Acme a link to connect their bank’s check images.”
  • ledgersync_create_connect_link takes a new setting, for_check_images. That link lists only the banks whose check images LedgerSync can connect, and connects them through LedgerSync’s own connection. Before, a link for banks like Regions or Bank of America always used the bank’s usual route, which has no check images.
  • The check images show in the LedgerSync app. A check images link brings no live transaction feed, so send a normal link too for the bank’s transactions.
  • When a bank in its answer gives check images, ledgersync_search_institutions now adds a check_images_note on how to connect them.
  • The AI now says exactly where the bank sign-in happens: on LedgerSync’s page, or on our bank-data partner’s secure page for some banks. Never in the chat.
  • Claude and ChatGPT keep a copy of LedgerSync’s tools. To get the new setting, reconnect in Claude or click Refresh in ChatGPT, then start a new chat. See After a LedgerSync update.
AI connector

Teammates can use LedgerSync in Claude and ChatGPT

Administrators and employees at a firm now connect LedgerSync in Claude or ChatGPT with their own LedgerSync login.
  • Each person sees only what they see in the LedgerSync app: an employee sees the clients assigned to them, an administrator sees the whole firm.
  • A client an employee creates through the AI is assigned to them, as in the app.
  • The firm owner gets an email each time a teammate connects.
  • Changing or resetting a password now disconnects only that person’s AI connections. The firm owner’s password change no longer disconnects teammates.
  • Sub-account (firm) logins and their employees, and logins with access days and hours, cannot connect yet. Client logins and support sessions cannot connect.
  • Only the firm owner links the desktop app. Teammates can check whether it is connected.
See Use LedgerSync in Claude or ChatGPT.
AI connector

A simpler AI connector guide for LedgerSync customers

The AI connector pages are now written for firms that already use LedgerSync.
  • A firm new to LedgerSync signs up in the LedgerSync app first (ledgersyncappv2.com/sign-up), then connects Claude or ChatGPT.
  • The setup pages keep their title and progress bar at the top while you scroll, and the setup videos open over the page.
  • The Claude Team or Enterprise and ChatGPT Business, Enterprise or Edu steps are now at the top of each setup page.
See Use LedgerSync in Claude or ChatGPT.
AI connector

Sign in to the AI connector with your LedgerSync app login

When you connect LedgerSync in Claude or ChatGPT, the LedgerSync sign-in page now opens.
  • Firms that use the LedgerSync app sign in with their normal LedgerSync username or email and password, plus their two-step code if they use one. No developer portal account and no email to support.
  • New firms create their LedgerSync account on that page: Create account, the sign-up form, then the order page. Our team reviews each new firm once. Until then the connector answers the new code firm_under_review when asked to connect a bank. Clients can be added.
  • Only the firm owner can connect for now.
  • Approved developer accounts type their developer portal email on the LedgerSync page, which sends them to the developer portal sign-in. An application still waiting for its first approval cannot connect yet.
  • Connections made before this change keep working.
  • The portal’s AI sign-up page (/start/ai) now sends new firms to the LedgerSync app sign-up.
See Use LedgerSync in Claude or ChatGPT.
AI connectorInstitutionsConnections
A few banks that LedgerSync reaches through FDE need the LedgerSync desktop app for refreshes. The app runs on a computer at the firm. The API and the AI connector now say which ones, and the connector can link the app.New field requires_desktop_app. It is optional, so it is additive for existing clients.
  • On institutions (GET /v3/institutions and GET /v3/institutions/{id}): true when LedgerSync reaches the bank through FDE and needs the desktop app to refresh it. Omitted when unknown.
  • On connections (GET /v3/connections/{id} and GET /v3/clients/{client_id}/connections): on FDE connections only, omitted for other sources. When true, the desktop app must be open and connected on the firm’s computer for a refresh to run.
Two new AI connector tools, on the live connector only:
  • ledgersync_get_desktop_app_status says whether the desktop app is linked and connected right now, with the Windows and Mac download links and the setup steps.
  • ledgersync_pair_desktop_app links the app with the 7-character code it shows. It replaces any computer linked before, so the AI asks the user to confirm first. Each account gets 10 tries an hour and 20 a day.
  • Bank search results and connections in the connector mark the banks that need the app.
  • New codes for linking: pairing_code_invalid, desktop_app_session_closed, desktop_service_unavailable and pairing_rate_limited.
Refreshes stopped by the desktop app say so.
  • A refresh FDE stops because the bank needs the desktop app and the app is not connected fails at once with desktop_app_required, and the connection reads requires_action with the same code. A refresh stopped while waiting for the choice of whether to use the desktop app fails with desktop_app_route_choice. Neither reached the bank, so neither counts against the daily refresh budget.
  • When FDE asks at the start of a refresh whether to use the desktop app, a working FDE connection no longer shows requires_action with mfa_required for that moment. It stays active, and no webhook fires for the question.
See Errors and Banks that need the LedgerSync desktop app.
AI connector

New firms can sign up for Claude and ChatGPT in a few minutes

A firm new to LedgerSync can now sign up with only its name and a card at portal.ledgersyncappv2.com/start/ai/new. There is no sandbox step and no developer form.
  • The first charge is one month after the card is added.
  • We review each firm once, within 2 business days, and email the answer. A sign-up we do not approve, or one that closes before its review, is not charged.
  • The firm can add LedgerSync to Claude or ChatGPT while it waits; it starts working as soon as the firm is approved.
  • On the live connector, Claude and ChatGPT now explain in plain words what a new firm still needs to do, with a link to the right page.
  • Developers building on the API keep the usual path: a sandbox request, then the Live application and payment.
See Use LedgerSync in Claude or ChatGPT.
AI connector

Use LedgerSync in Claude and ChatGPT

Add the LedgerSync connector to Claude or ChatGPT and ask in plain words: find a client, send them a secure link to connect a bank, read balances, transactions and statements, see which connections need attention, and refresh a connection. Bank sign-ins stay on LedgerSync’s own pages, never in the chat.
  • Live: https://api.ledgersyncappv2.com/mcp. Sandbox: https://api-sandbox.ledgersyncappv2.com/mcp.
  • Sign in with your developer portal account. Live needs the same approval and payment as live API keys; existing LedgerSync app customers are linked, not charged again.
  • The portal’s new Connected apps page lists the AI apps connected to your account and disconnects one at once.
See Use LedgerSync in Claude or ChatGPT.
ConnectionsWebhooks

Refreshes of MX and FDE connections now finish

POST /v3/connections/{id}/refresh on an MX or FDE connection now ends the way it does on Finicity: the operation turns succeeded or failed once every account on the connection has reported, and account.refresh.completed or account.refresh.failed fires for each account. The one exception is an FDE refresh refused straight away (below), which fails the operation without a webhook.
  • MX and FDE refresh operations reach a real verdict. They used to record no accounts to wait on, so every one ended failed with refresh_timed_out after 90 minutes, whatever happened at the bank.
  • MX refresh webhooks are sent for client connections. account.refresh.completed and account.refresh.failed were never sent for an MX connection that belongs to a client, which is every MX connection made through the API. They are now.
  • An MX refresh that finds nothing new still finishes. MX sends LedgerSync no notice when a sync brings no new or changed transactions, so LedgerSync now asks MX once the sync is over and reports it. Expect the verdict about three minutes after the call.
  • An FDE refresh the bank answers with a security question fails at once with mfa_required, instead of waiting out the timeout. Send the user a fix link from POST /v3/connections/{id}/reauthorize. Any other refusal fails at once with source_internal_error, with the FDE status in source_diagnostic_code (for example FDE-UPDATED_ERROR). No webhook fires for either; read the operation.
  • FDE connections report last_refreshed_at. It was always null on FDE connections. Their created_at and updated_at, which follow it, fill in too.
  • Sandbox refresh operations finish too. On sandbox they never matched their results and always timed out.
  • An earlier sync can no longer answer your refresh. A sync result that reached LedgerSync before your call, but was still being processed, could close your refresh with that older data. Only a result that arrives after your call answers it now.
Correction to the 2026-08-25 entry. It said MX and FDE connections “already reported completion correctly”. They did not: their refresh operations always ended refresh_timed_out, as described above.See Keeping data fresh.
Statements

Statement downloads work again

From 2026-09-24 until the fix on 2026-09-25, GET /v3/statements/{id}/download answered 500 instead of the PDF. It returns the file again. If a scheduled download failed in that window, run it again.
ConnectionsWebhooks

Connection status and webhooks now agree

GET /v3/connections/{id} and the connection.* webhooks now report the same status for the same state.
  • MX connections read their real status. The read endpoint used to report any unhealthy MX connection as disconnected. It now follows MX’s own member status, the same one the webhooks report: MX PREVENTED (too many failed sign-ins) and DENIED read requires_action with credentials_invalid, MX DISCONNECTED and CLOSED read disconnected, and a healthy or still-syncing member reads active. Expect some MX connections to move from disconnected to their real status.
  • The reason matches the status. A connection.requires_action for MX PREVENTED used to carry institution_unavailable with is_user_actionable: false. It now carries credentials_invalid, actionable. An MX-locked account is the one deliberate exception: requires_action with account_locked_by_institution and is_user_actionable: false, because the user acts at the bank. See Errors for the MX codes behind each error.
  • Connection events name the client. data.connection.client_id is now present on status-change events whenever the owner can be resolved, and the key is omitted, never null, when it cannot.
  • Deleting a connection sends connection.disconnected, as the testing guide already described.
  • Status changes found by the scheduled daily refresh now send webhooks. A connection that recovers on its own after connection.failed sends connection.active. One exception remains: a Finicity MFA challenge on an existing connection reads requires_action on GET without an event.
  • A new MX connection is named by a webhook. When a first MX link connects, or the bank rejects it, a connection.* event now carries the canonical con_MX_... id. Before, only the initiate connection.requires_action under the placeholder id fired, so an MX link that connected without an MFA step, or failed outright, had to be found by listing the client’s connections.
disconnected no longer means “the user revoked access”: it means the source stopped aggregating the connection, you deleted it, or (for Finicity and portal connections) LedgerSync’s own refresh of it failed. See Connection lifecycle.
AccountsTransactions

Card terms, account masks and retracted transactions

  • liabilities appears only on liability accounts, such as credit cards, lines of credit, loans and mortgages. It no longer appears on checking, savings or investment accounts, even when the bank reports a figure there. Identify cards by type: credit_card; loans and mortgages carry the object too.
  • MX now reports the statement balance. It was dropped before, so every MX statement_balance was null. It fills in as each MX account next syncs.
  • Card terms refresh on the scheduled daily refresh, not only when a refresh is triggered.
  • mask is now filled for MX, portal (FDE) and uploaded-statement accounts, as at most the last 4 digits of the account number. It used to be null for all three.
  • GET /v3/transactions/{id} returns 404 for a Finicity transaction the source retracted. The list never returned those; the single read now agrees. If you stored such an id, treat the 404 as “removed”.
Corrections to the 2026-09-04 entry. The object’s presence is not the card signal (loans and mortgages carry it). Finicity does not report all five fields everywhere, and MX does report the statement balance. The real coverage, measured per source and field, is in Credit-card terms.
Webhooks

Webhook subscriptions now report delivery health

last_success_at and last_failure_at were in the WebhookSubscription schema but documented as reserved, and never appeared in a response. They are now populated and returned on every GET, LIST and PATCH:
Read them together: last_failure_at newer than last_success_at means deliveries are failing right now, and that is the condition worth alerting on. A subscription stays active however badly delivery is going, because we keep retrying rather than disabling your endpoint, so status alone was never enough to tell you your webhooks had stopped arriving.Failures are stamped on every failed attempt, including retries still to come, and a later success never clears last_failure_at, so an endpoint that keeps flapping stays visible. Both fields are omitted rather than sent as null when there is nothing to report, so check for the key rather than comparing to null.Existing subscriptions were backfilled from delivery history, so the timestamps are meaningful immediately rather than only reflecting deliveries from today onward. See Check delivery health.
Accounts

Credit-card terms on the Account

Credit limit, statement balance, minimum payment and payment due date were not available on the API. They are now, on a new liabilities object:
The object is absent unless we know something, so it is null for every account that is not a credit card or line of credit, and null for cards at banks that publish no terms. Check for the object before reading fields inside it, and treat an absent field as “the bank did not say”, not zero.Bank connections report all five fields where the bank publishes them. MX reports every field except the statement balance, which it does not track. Portal connections and uploaded statements report none.statement_balance is signed like current_balance: what the client owes is negative, so the two agree on the same account. That sign is normalized rather than reported, because banks disagree with each other about it. One consequence: a card that was in credit at statement close is reported as owed. credit_limit, available_credit and minimum_payment are always positive.payment_due_date is only as current as the last refresh, so a date in the past means the cycle rolled over before our last pull, not that a payment is overdue. See Credit-card terms.
TransactionsAccounts

Fetching a single transaction works on every source, and accounts now say how far back they go

GET /v3/transactions/{id} resolves ids from all four sources. It only ever worked for Finicity and uploaded-statement transactions. An MX id came back 404, and a portal-connection (FDE) id failed outright, even though both were listed by GET /accounts/{id}/transactions moments earlier. An id we hand you now resolves.FDE transactions carry account_id. The field is documented as required and was arriving null on that source, so there was no way to tell which account a transaction belonged to without tracking the request you made to get it.New oldest_transaction_date on the Account. It answers “how much history do I have here”, per account:
It is measured from the transactions we hold rather than copied from what the upstream provider reports about itself, so it is exactly the minimum date you get if you page that account’s whole history. It is null when we hold no transactions for the account yet, and it moves backwards if a backfill lands earlier data, so read it rather than caching it. See How far back does the history go?.The field is optional and nullable, so it is additive for existing clients.
Accounts

Account type is now guaranteed, and balances are signed consistently

Three things about the Account object were true in the reference but not in the responses. They are true now, and a contract test keeps them that way.type is always one of the six documented values, and never null. It used to be a pass-through of whatever the underlying aggregator called the account, so a Finicity credit card arrived as creditcard, an uploaded-statement one as credit card, and only MX ever sent the documented credit_card. Accounts from portal-based connections sent no type at all. If you branch on type === "credit_card" you will now match every credit card, on every source.New subtype field. The coarse enum discards real detail, so the source’s own label now travels alongside it: money_market, mortgage, line_of_credit, ira, plan_401k, bank, and so on. It is an open vocabulary rather than an enum, so new values are never a breaking change. Branch on type; use subtype for display or for finer routing you control.If you were relying on the old raw values, subtype is what you were reading before, so switching your comparison from type to subtype gets you exactly the previous behaviour.current_balance is signed from the account holder’s point of view. Positive is value the client holds, negative is value they owe, on every source. Previously the sign followed whichever aggregator supplied the account, so the same credit card could be positive or negative depending on how it happened to be connected. Summing current_balance across a client’s accounts now gives net worth directly.A real credit balance on an overpaid card is still positive, and a checking account in overdraft is still negative, so the sign stays meaningful in both directions. available_balance is unchanged: on a card it is available credit, not a debt, and it is never sign-flipped.The account.refresh.completed webhook is signed the same way, so the value you read from GET /accounts and the value you receive on the event agree for the same account.available_balance capability now reports honestly. It was declared never for every source while MX accounts were returning a real value. MX is now sometimes, which is accurate: MX reports it for many checking accounts and few credit cards. Every other source is genuinely never.Also fixed: the transactions endpoint told you to check a status field that does not exist. The field is the pending boolean.
Connections

Refresh now pulls transactions, and its operation actually finishes

POST /v3/connections/{id}/refresh on a Finicity connection re-aggregated balances only. Transactions were not pulled, last_refreshed_at never moved, and no account.refresh.completed webhook fired - so the operation it handed you stayed queued forever and polling it never returned a terminal state. Fixed. The call also answers in milliseconds now instead of blocking.
  • Transactions are pulled and last_refreshed_at advances, so re-reading GET /v3/accounts after the webhook shows what you would expect.
  • The operation reaches a terminal state. A refresh is one aggregation at the bank but reports per sub-account, so the operation turns succeeded or failed once every account on the connection has reported - and resolves on its own if the events never arrive, so polling always terminates.
  • Partial failures are reported as failures, not as partial successes. If the aggregation ran but one account did not come back, you get refresh_partially_failed with a per-account breakdown in error.accounts[], so branching on status alone cannot quietly ship a connection that is missing an account’s transactions.
  • New error codes: refresh_partially_failed, refresh_timed_out, refresh_outcome_not_observable, no_refreshable_accounts.
  • Refresh now has a daily budget separate from the request-rate limit, since each one is a real aggregation at the institution. Over it, the call returns 429 rate_limit_exceeded without touching the bank.
  • estimated_seconds on the 202 was raised from 10 to 120 to match how long a refresh actually takes. It is a best-effort floor, not a deadline.
MX and FDE connections are unchanged - they already reported completion correctly and keep the behaviour they had.See Keeping data fresh and the connection refresh error codes.
Connections

Reauthorize (fix) a broken connection in place

New POST /v3/connections/{connection_id}/reauthorize repairs a connection that needs credentials or consent renewed, aiming to preserve existing IDs.
  • Returns a reauthorize action with a hosted reauth_url; redirect the end user there to re-enter credentials or re-consent at the institution, then wait for connection.active (or connection.failed).
  • ID preservation depends on the provider and whether the repaired login and accounts match the stored connection. Starting another Finicity Full Connect session can also reuse an existing bank login; use that flow to add accounts, and verify canonical IDs afterward.
  • Available for FINICITY, MX, and FDE. Uploaded-statement (PDF) sources are read-only and have no connection to reauthorize.
See Connection lifecycle.
ChecksFDE

Check images are now readable from the API

The check_images capability has had a read path since day one on the data side, but no route to fetch it. Three endpoints now serve it, nested under the owning account:
  • Requires the new read:checks scope. Existing keys don’t have it — tick it on a new key in the portal, or PATCH /v3/api-keys/{id} to widen the key you already have without rotating its secret.
  • FDE-only. Non-FDE accounts return an empty list rather than an error, so you can call this uniformly across sources.
  • There is no back image — LedgerSync captures the front of a check only, so there is no side field.
  • Bank-reported fields and ocr.* are kept separate, so you can always tell what the bank said from what OCR read off the paper.
  • The sandbox Ledgersync Bank (ins_a7397a8d0656e1b7) returns real check images, so the whole flow is testable end-to-end.
See Check images.
ConnectionsMX

On-demand refresh now works for MX connections

POST /v3/connections/{id}/refresh now supports MX (con_MX_*) the same way it already supported Finicity and FDE: it returns 202 Accepted with an operation_id and triggers a fresh aggregation. (Previously MX returned 502 refresh_not_supported.)
  • Completion is asynchronous — listen for the account.refresh.completed webhook (or account.refresh.failed on terminal failure), then re-read GET /v3/accounts for the updated balances and transactions.
  • All three connectable sources (Finicity, MX, FDE) now answer /refresh uniformly, so you no longer need to special-case MX.
AccountsTransactions

PDF: a read-only fourth data source

Accounts created from uploaded bank statements now surface through the API. Their accounts and transactions come back with source: "PDF" (ids like acc_PDF_42, txn_PDF_8837).
  • Read-only: PDF isn’t linked through the widget and never appears under GET /v3/connections — a PDF account’s connection_id is null.
  • Exposed on GET /v3/accounts and GET /v3/accounts/{id}/transactions. Statements aren’t served for PDF accounts.
  • Everything else — pagination, filtering, the response shape — reads exactly like an aggregator source.
See the data model.
Pagination

Cursor pagination on all list endpoints

Every list endpoint now returns an opaque next_cursor and a has_more flag. Pass next_cursor back as the cursor query param and loop until has_more is false.
  • Keyset (seek) based, so it stays correct even as rows are added or removed between pages — no skipped or double-counted items.
  • limit defaults to 100 (max 500; higher values are capped, not rejected).
  • Cursors are opaque and query-scoped — reusing one against a different account, filter, or date window returns 400 invalid_request.
Full guide: Pagination.
Statements

Extract transactions from a bank-statement PDF

New POST /v3/statements/extract — upload a bank-statement PDF and get back the OCR-extracted transactions, powered by the same engine behind the LedgerSync app’s Bank Statement Converter.
  • Customer-level and standalone: no client_id, and the upload never attaches to a connection, account, or stored statement.
  • Async: returns 202 with an operation_id; poll GET /v3/operations/{id} (most statements finish in under a minute). The result carries kind: statement_extraction. There’s no webhook for this one — polling is the completion signal.
  • PDF only, up to 30 MB (larger uploads are rejected with 413). Requires the write:statements scope. Per-customer concurrency and daily budgets apply on top of the standard rate tiers — exceeding them returns 429.