MCP App Store
by KeenEthics
YESbusiness icon

YESbusiness MCP Server for ChatGPT

By YESbusiness Sàrl

Overview

YESbusiness is bookkeeping and VAT software for Swiss small businesses, freelancers and associations, connected to the user's bank. The YESbusiness app links to Swiss banks through EBICS (more than 50 banks, including UBS, PostFinance, Raiffeisen and the cantonal banks) and imports the account's transactions automatically. This plugin lets the user work on those bank transactions and the rest of their own books from ChatGPT: see what is still unbooked and match it to receipts; propose bookings the user applies with one click, or book directly within an amount cap; read balances, entries, VAT statements and reports and export them; import bank statement files (camt.053) and spreadsheets of entries; apply booking rules; and manage accounts, VAT codes, VAT periods, cost centres and accounting years. Access uses OAuth with the user's existing YESbusiness login. On the consent screen the user chooses the companies and one of three rights levels — read only, booking with the user's confirmation, or autonomous booking within an amount cap the user sets — and whether receipt files may be read. Every entry an assistant writes is tagged, listed on the company's AI activity page and can be undone. Limitations: the bank connection itself is set up in the YESbusiness app — the plugin never handles bank credentials, payments, subscriptions or members, never moves money, never deletes data, cannot book in a closed accounting year, and nothing is sold through it.

How to connect

  1. ChatGPT: Open the YESbusiness listing in the ChatGPT app directory and choose Connect. ChatGPT listing YESbusiness exposes 51 tools here.

Tools

apply_booking_rules

ChatGPT
Runs the company's booking rules over every unbooked transaction of one cash account — the Banking page's 'Apply rules' — booking each transaction a rule matches. On a connection that proposes (write:confirm) this returns a PREVIEW: how many transactions wait, how many rules exist, and the deep link to the Banking page where the user applies them (the result cannot be proposed yet); an autonomous connection (write:auto) runs the rules and returns the summary (booked, skipped and why). Rules write complete bookings with the rule as their source, so they are not undone by revert_connector_writes.

apply_proposal

ChatGPT
Applies a pending proposal — the draft becomes a complete booking (or the stored batch is written). It REQUIRES the proposal's confirmation token, which only the YESbusiness card in the host's UI receives: a model never has it and must not ask the user for it. Without the token the call is refused before anything happens; the user applies from the deep link instead. Only the user who received the proposal can apply it, and only before it expires.

archive_account

ChatGPT
Archives an account: it stays on its old bookings and in the reports but takes no new bookings and leaves the pickers. Accounting data is never deleted; the user can unarchive it in the settings. Refused by the backend while the account is still in use where it must not be (a cash account with a bank connection, a VAT account). On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

archive_cost_center

ChatGPT
Archives a cost center: it stays on old bookings and leaves the pickers. Never deleted; the user can restore it in the settings. On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

archive_vat_code

ChatGPT
Takes a VAT code out of the pickers (is_usable false): existing bookings keep it, new ones no longer see it. VAT codes are never deleted here; the user can make it usable again in the settings. Use it when a rate ended (together with a new code for the new rate). On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

attach_document_to_booking

ChatGPT
Attaches an uploaded document to an existing booking (adds it to the booking's attachments; nothing else about the booking changes). The booking is re-saved through the same validation the panel uses, and refused if the round trip would alter its figures. Bank reconciliation bookings cannot be changed here — the user attaches on the Banking page, or you pass documents_ids when proposing the bank transaction's booking. Needs the booking's id (list_bookings, get_journal) and the document's id (upload_document).

book_bank_transaction

ChatGPT
Books ONE imported bank transaction DIRECTLY (reconciles it) — only on a connection the user granted autonomous writes (write:auto), within its amount cap and in an open accounting year; otherwise refused, and propose_bank_transaction_booking is the way. The bank leg (amount, side and date) comes from the transaction itself — you name the cash account the transaction sits on and the counter account; money in debits the cash account and credits the counter account, money out the reverse. Transactions in a foreign currency need the Banking page. Attach the invoice or receipt the transaction pays with documents_ids — a bank booking cannot get a document afterwards through attach_document_to_booking. The booking is validated exactly as the banking screen's inline booking, written complete with the connector as its source, and can be undone with revert_connector_writes. Before calling this tool, show the user the booking you are about to prepare (date, description, every leg with account and amount) and get their agreement; the tool then validates it and returns the resolved booking exactly as it would be saved — show that too. Money rules: every amount is an INTEGER in CENTS of the company's HOME currency (read it with get_bookkeeping_guide or get_currency_settings first — never assume CHF); 12.50 is 1250. Account numbers, VAT codes and cost centers are the company's own — read them (list_accounts, list_vat_codes, get_cost_centers) before naming one. Reports are pre-summed: never add child rows yourself.

create_account

ChatGPT
Adds an account (or a group header) to the company's chart of accounts. The number must be unique and follow the chart's numbering; the parent is a group of the same statement. account_type_id: 1 cash and bank, 2 other assets, 3 liabilities, 4 expenses, 5 revenue, 6 equity, 7 group (a header that takes no bookings). default_vat_code is one of the company's VAT codes, applied to new bookings on the account; currency_id (from list_available_currencies) only for a foreign-currency account. The values are checked by the backend before anything is written. On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

create_accounting_year

ChatGPT
Creates an accounting year (needed before anything can be booked in its dates). An accounting year has a name, inclusive dates that must not overlap another year, a status (open takes bookings; blocked and revised are closed) and its booking-number format: prefix + the next number zero-padded to counter_size. Read get_currency_settings for the existing years first. On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

create_booking

ChatGPT
Writes a booking DIRECTLY, without the user's click — only on a connection the user granted autonomous writes (write:auto), within its amount cap and in an open accounting year; otherwise refused with the reason, and propose_booking is the way. The legs are validated exactly as the booking panel validates a save, then the booking is written complete (status 'draft' keeps it provisional), tagged with the connector as its source, so it appears in the AI activity list and can be undone with revert_connector_writes. Before calling this tool, show the user the booking you are about to prepare (date, description, every leg with account and amount) and get their agreement; the tool then validates it and returns the resolved booking exactly as it would be saved — show that too. Returns the booking number, id, the resolved booking and a deep link. Money rules: every amount is an INTEGER in CENTS of the company's HOME currency (read it with get_bookkeeping_guide or get_currency_settings first — never assume CHF); 12.50 is 1250. Account numbers, VAT codes and cost centers are the company's own — read them (list_accounts, list_vat_codes, get_cost_centers) before naming one. Reports are pre-summed: never add child rows yourself.

create_cost_center

ChatGPT
Adds a cost center to one of the company's two families (family 1 and 2 — the company's own axes, e.g. Department and Project; get_cost_centers shows their titles). Cost centers must be activated on the company first. On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

create_vat_code

ChatGPT
Adds a VAT code to the company (the code a booking leg carries to have its VAT split to the VAT accounts). rate is a FRACTION (0.081 for 8.1%, 0.026, 0.038, 0 for exempt). kind 'output' is VAT on sales (revenue), 'input' is deductible VAT on purchases. booked_in_account_number is the VAT account the tax is posted to (2200 VAT due for output, 1170 input VAT); entered_net says how a booking's amount is read: true = the amount EXCLUDES the VAT (it is added on top), false = the amount INCLUDES it (gross, the usual case); gross_in_account_number is required for a gross code. effective false marks a net tax rate (TDFN / Saldosteuersatz) code; vat_only a code that books the VAT alone (import VAT). report_codes are the AFC form lines (e.g. '302' for 8.1% turnover, '400' for input tax on goods), comma-separated. Read list_vat_codes first: the standard codes usually already exist. On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

create_vat_period

ChatGPT
Adds a VAT period. A VAT period says, from a date, whether the company is VAT-registered, its VAT number, the method ('effective' or 'net' tax rate) and the basis ('agreed' = on invoicing, 'collected' = on payment). Periods must not overlap; date_until empty = open-ended. A change of method or basis is a new period starting on its first day. The dates and overlap are checked before anything is written. On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

export_report

ChatGPT
Produces a downloadable FILE of a report and returns its link: the balance sheet, the income statement or the account statement as CSV, XLS (a real .xlsx workbook) or PDF, and the journal (every leg of every booking in the period, chronological) as PDF only. Use it when the user wants the report itself to download, print, send or open in a spreadsheet; for the figures, call the matching get_* tool instead. date_to is the balance sheet's report date and every other report's period end; date_from is required for the three period reports. The account statement takes the same scope / account inputs as get_account_statement. The result is {status, url, file_name}: give the user the url as a link to click. The link is signed and short-lived — hours, not days — so do not store or repeat it later; if it has expired, export again. status 'processing' means the statement was too large to build in one request: it is prepared in the background, the download link is e-mailed to the user and it also appears under their personal settings, My exports. Amounts in the file are in the company's home currency.

get_account_statement

ChatGPT
The account statement: the movements of one or more accounts over a date range, as a flat list of ledger rows with a running balance — this is the tool for 'what happened on account X', not for totals. Choose the scope: 'single' with account (the account NUMBER, e.g. '1020'), 'range' with from_account and to_account (account numbers), or 'all' for every account. Each row has the booking date, number and status, the description, the counterpart accounts, and debit / credit / running as integers in CENTS of the home currency — divide by 100 and present francs with two decimals. Rows with is_opening true are the account's starting balance, not a movement. Only the FIRST page of rows is returned: items_total is how many movements match in total, items_shown how many you can see. When items_total is larger than items_shown, say so and offer to narrow the dates instead of summing the rows you got — for totals, use the balance sheet or income statement.

get_analysis

ChatGPT
One call for an analysis over a date range: the NET MOVEMENT per account per month or quarter (group_by 'account'), or the booking totals per counterparty per period (group_by 'counterparty' — the counterparty is the booking's description, so recurring items like a landlord, a telecom provider or a customer group together). Use it for trends, seasonality, 'which months were expensive', 'how much did we pay X over the year' and 'top counterparties'; for a statement's official figures use get_income_statement / get_balance_sheet. Only complete bookings count. The result is {home_currency, periods, rows}: periods are labels ('2026-01' or '2026-Q1', oldest first), each row has one cell per period in the same order plus total_cents, all integers in CENTS of the home currency — divide by 100. For accounts, a cell is debits minus credits in the period: positive means the account grew on its debit side (an expense incurred, an asset acquired), negative on its credit side (revenue earned, a liability taken on). Rows are sorted by total, largest absolute amount first, capped at 100 (rows_truncated says if more existed). cost_center_id restricts the figures to the legs carrying that cost center (ids from get_cost_centers).

get_balance_sheet

ChatGPT
The balance sheet: what the company owns and owes as balances up to one date (date_until), as a nested tree of account groups and accounts. Sign convention in the raw values: assets are POSITIVE, liabilities and equity NEGATIVE — that is how the report is built, not an error, so talk about amounts held or owed instead of reading a minus sign out loud. The passive side also carries three synthetic rows, identified by a string id: 'yearly_result' (the provisional result of the open accounting year(s), negated like the rest of the passive side, so a profit shows as a negative value), 'currency_revaluation' (unbooked foreign-currency revaluation) and 'total_passive' (liabilities and equity including those results). Set compare_date to add the second period and the difference. The tree is PRE-SUMMED by the server: a group row's values already include every account under it, so read a total off the group row itself and never add child rows up yourself. Every amount is an integer in CENTS of the company's home currency — divide by 100 and present francs with two decimals, never the raw integer. Each row has values.period_1 (the main period), plus values.period_2 and values.difference (period_1 - period_2) only when compare dates were sent.

get_bookkeeping_guide

ChatGPT
The Swiss bookkeeping conventions this company's books follow, as text to read before interpreting figures or preparing a booking: account numbering and types, what a draft is and the booking rules, VAT codes, rates and methods, and the money rules (integer cents, home currency, sign conventions). It also returns the company's own settings the conventions depend on: home currency, activated foreign currencies, accounting years with status and exchange-rate method, and whether the company is VAT-registered. Static per company; call it once per conversation, not per question. Choose a topic to get one section, or omit it for all four.

get_cost_centers

ChatGPT
The company's cost centers: the two families (the company's own axes, e.g. Department and Project) and the cost centers inside each. A family with an empty title is one the company does not use. Archived cost centers are excluded — they stay on old bookings but are never offered for new work. Call this before answering anything about cost centers, departments or projects, or before naming one in a booking. An empty list means the company has not set any up; whether the feature is on at all is cost_centers_activated on the company.

get_currency_settings

ChatGPT
The company's foreign-currency setup: whether foreign currencies are activated at all, the home currency, the list of activated currencies, and the accounting years with each year's exchange-rate method and status. Call this before answering any question about using, adding or converting currencies.

get_document

ChatGPT
One document of the company — a receipt, an invoice, a statement attached to a booking — as its metadata plus a way to read it: {id, name, mime, size_bytes, signed_url, base64?}. signed_url is a short-lived (about an hour) link to the file; base64 is the file itself and is only returned when inline is true AND the file is 2 MB or smaller. document_id comes from a booking's attached_document_ids (list_bookings, get_journal) or from list_unattached_documents. Only available when the connection was granted access to files; every call is logged with the document id. Never quote account numbers or IBANs you read in a document back in full.

get_income_statement

ChatGPT
The income statement (profit and loss): revenue and expenses over a date range (date_from to date_to, inclusive), as a nested tree of account groups and accounts. Sign convention in the raw values: revenue is POSITIVE and expenses NEGATIVE, so a group total is revenue plus (negative) expenses and a positive top-level result means a profit, a negative one a loss. Every row is a real account or group — no synthetic rows. Set both compare_from and compare_to to add the compare period and the difference. The tree is PRE-SUMMED by the server: a group row's values already include every account under it, so read a total off the group row itself and never add child rows up yourself. Every amount is an integer in CENTS of the company's home currency — divide by 100 and present francs with two decimals, never the raw integer. Each row has values.period_1 (the main period), plus values.period_2 and values.difference (period_1 - period_2) only when compare dates were sent.

get_journal

ChatGPT
The journal: the company's bookings in CHRONOLOGICAL order (oldest first) over a date range, one page at a time — the view for 'what was booked in March', 'walk me through last week's entries' or checking a sequence of booking numbers. date_from and date_to are required (YYYY-MM-DD, inclusive). Each row: date, number, status (draft = provisional, not yet in the reports; complete = booked), description, the debit and credit account labels, amount_home (integer CENTS of the home currency — divide by 100), the cost centers on its legs and its source (how it was created: a bank import, the assistant, a connector). Filters: status, search (description or number), account_id (only bookings with a leg on that account). items_total says how many bookings match; ask for the next page only when the user needs it, and never add amounts across many rows yourself — use the income statement or balance sheet for totals. For the leg-by-leg journal document (every debit and credit line), call export_report with report 'journal'.

get_monthly_flows

ChatGPT
Cash in and cash out per MONTH and per cash account (bank, cash, payment-provider balances) over a date range — the Banking screen's chart as data. Use it for 'how much came in and went out each month', runway and seasonality questions about cash; for profit use get_income_statement, for balances get_balance_sheet. date_from and date_to are YYYY-MM-DD, inclusive; one bucket per month in the range, oldest first, empty months included. inflow / outflow are positive integers in CENTS of the home currency — divide by 100; net is inflow minus outflow. Counted: the legs of complete bookings on the cash accounts PLUS imported bank transactions not yet booked (so the figures match the bank, not only the ledger). An account with no movement in a month is absent from that month.

get_proposal_token

ChatGPT
For the YESbusiness card only: returns the confirmation token of a proposal this connection created, so the card can apply it when the token did not arrive with the proposal. Not for the model: a model must never apply a proposal.

get_setup_status

ChatGPT
The company's setup state: whether its accounting exists yet (has_accounting), the accounting's status and VAT registration, the chart of accounts, and the 14-day trial (started / ends_at / expired). Call this when the user asks about setting up or starting their accounting, why the dashboard shows the start-accounting invitation, or their trial. While has_accounting is false, the way forward is the new_accounting page (ui_navigate), and no bookings or reports exist yet.

get_vat_statement

ChatGPT
The VAT statement (the AFC declaration) for a date range: one row per numbered form line, each with the line's own figures. method says which form it is — 'effective' (the ordinary method, with an input-tax deduction) or 'net' (the net tax debt rate, TDFN / Saldosteuersatz, where a flat rate on turnover replaces the input tax and the input-tax lines do not exist at all). Use it to answer what is owed for a period, or which method the company is on. rate is a FRACTION (0.081 is 8.1%) and is the saldo rate on the net form's turnover lines. net / vat / gross are integers in CENTS of the home currency — divide by 100. A null figure means the line has none, which is not the same as zero. The headline lines are 299 (total taxable turnover), 399 (total tax due), 479 (total input tax, effective only) and 500 / 510 (payable, or credit in the company's favour). Refused when the range is not covered by a VAT period, or crosses a change of method — each method has its own form, so ask for one period per method.

import_bank_file

ChatGPT
Imports a bank statement file (camt.053 / camt.054 XML, a bank's CSV, or a ZIP of them) onto one cash account: the statement lines become bank transactions waiting to be booked, and the company's booking rules are applied to them at once. On a connection that proposes (write:confirm) this only PREVIEWS the import — file, size, target account, how many transactions are already waiting — and gives the deep link to the Banking page where the user imports; an autonomous connection (write:auto) imports and returns the result and the new count of unbooked transactions (a ZIP is unpacked in the background: status 'processing'). The file is sent as base64, 8 MB at most. Never import a file the user did not hand you for this account.

import_bookings

ChatGPT
Imports a spreadsheet of bookings (the YESbusiness booking template: date, description, debit account, credit account, amount incl. VAT, VAT code, currency, an optional number grouping rows into one booking). Every row is VALIDATED first exactly as the app's import validates — unknown accounts, unusable VAT codes, a closed year, an unbalanced group — and the errors come back by row for you to show the user; nothing is written while any row fails. Then: on an autonomous connection (write:auto) the rows are imported, one complete booking per row or group; on a connection that proposes (write:confirm) ONE proposal holds the whole validated set (home-currency rows only) and the user applies it from the deep link. rows is a list, one object per spreadsheet line: {"date": "2026-03-31", "description": "Rent March", "debit_account": "6000", "credit_account": "1020", "amount_cents": 150000, "vat_code": "V81", "number": "A1", "currency": "CHF"}. amount_cents is the amount INCLUDING VAT, an integer in cents; vat_code, number and currency are optional (an empty currency is the home currency). Rows sharing a number form ONE booking with several leg pairs and must share the date. 500 rows at most. Money rules: every amount is an INTEGER in CENTS of the company's HOME currency (read it with get_bookkeeping_guide or get_currency_settings first — never assume CHF); 12.50 is 1250. Account numbers, VAT codes and cost centers are the company's own — read them (list_accounts, list_vat_codes, get_cost_centers) before naming one. Reports are pre-summed: never add child rows yourself.

list_accounts

ChatGPT
The company's BOOKABLE accounts (its chart of accounts without group headers and archived accounts), sorted by number: id, number, name, account_types_id (1 cash and bank, 2 other assets, 3 liabilities, 4 expenses, 5 revenue, 6 equity, 8 year-end result), currencies_id (set on a foreign-currency account) and default_vat_codes_id. Read it before naming an account in a booking or a rule — numbers and names are the company's own, never assume one exists. Static per company; call it once per conversation.

list_alerts

ChatGPT
The announcements YESbusiness addressed to the user that they have not yet read or dealt with (newest first, 20 at most): the congratulations when an accounting is created, a notice the team sent, a step still to take. They belong to the USER, not to one company. Each has a heading, a body (prose in the user's language, may hold line breaks and '- ' bullets), an optional in-app link and a kind. Call it when the user asks whether anything needs their attention or what YESbusiness told them; you cannot dismiss one — the user does, on the dashboard.

list_available_currencies

ChatGPT
The catalog of currencies the company could activate (id, symbol, localized name). Use it to check whether a currency the user names exists and to get its exact symbol.

list_bookings

ChatGPT
The company's bookings (double-entry ledger entries), newest first, as one page with a total count. Filters: date_from / date_to (YYYY-MM-DD, inclusive, on the booking date — turn 'this month' etc. into an explicit range before calling); search (free text matched against a booking's description OR its number — good for a named item like 'rent', but it does NOT match account names or amounts); status. Each row has date, number, description, amount_home (integer, in cents of the company's home currency), currency symbol, and the debit/credit account labels. Use the returned total to answer whether or how many bookings match; do not sum amounts across many rows yourself. You CANNOT filter by income vs expense or by account here — if that is what the user wants, say you cannot look that up yet.

list_my_companies

ChatGPT
Lists the companies this connection may work on: the user's own memberships and, for a fiduciary, the client companies of their mandates — only those the user granted when connecting. Call this first: every other tool needs a company_id from this list. Never guess or reuse a company_id from another conversation. Each item: company_id, name, status, role_hierarchy (higher = more rights; 1 = read only), via_partner (true when access comes from a fiduciary mandate) and has_access (false = the company's subscription is not active, so its data cannot be read until it is).

list_open_receivables

ChatGPT
The company's OPEN customer invoices (accounts receivable): every invoice synced from EZYinvoice that still carries a balance — fully open or partially paid — soonest due first, one page at a time. Use it for 'who still owes us', overdue follow-ups and the receivables total. Each item: number, customer_name, issue_date, due_date, currency (the invoice's own), total_cents and open_cents in that currency as integers in CENTS — divide by 100 — and status ('open' or 'partially_paid'). Overdue = due_date before today. itemsTotal is how many open invoices exist; nextPage is null on the last page. You cannot record a payment or send a reminder here.

list_supported_ebics_banks

ChatGPT
The live catalog of banks YESbusiness can connect through EBICS, including the exact current count. Call it when the user asks whether a named bank is supported, which banks are supported, or how many are supported. An absent bank means only that it is not in the current EBICS catalog; never turn a failed lookup into an unsupported-bank claim.

list_unattached_documents

ChatGPT
The company's documents (receipts, supplier invoices, statements) that NO booking holds yet, newest first: id, file_name, mime, size_bytes, uploaded_on, external_link (true = a link to a file stored elsewhere, nothing to read). Use it when the user has put invoices into YESbusiness Documents and asks you to match them to bank transactions: read a candidate with get_document (its signed_url, not inline base64, unless the file is small), then pass its id in documents_ids on propose_bank_transaction_booking or propose_booking. search narrows by file name. Only available when the connection was granted access to files.

list_unbooked_bank_transactions

ChatGPT
The imported bank transactions that do NOT have a booking yet, on every cash account of the company (or on one, by cash_account_number), newest first. Use it to find the transaction a supplier invoice or a receipt pays: match the amount, the date (a payment usually follows the invoice date by days or weeks) and the counterparty name or the reference. Each item: id (the bank_transaction_id propose_bank_transaction_booking takes), cash_account_number, date, value_date, amount_cents (signed: negative = money out, positive = money in; integer CENTS in the transaction's currency — divide by 100), currency, description, counterparty, reference. unbooked_total per account says how many wait in all; when it is larger than what was returned, narrow with date_from / date_to or search. date_from and date_to are YYYY-MM-DD, inclusive.

list_vat_codes

ChatGPT
The company's USABLE VAT codes: id, code (what a booking leg carries, e.g. 'V81'), name, rate as a FRACTION (0.081 is 8.1%), kind ('output' = VAT on sales, 'input' = deductible VAT on purchases), entered_net (true = a booking's amount is entered WITHOUT the VAT, which is added on top; false = the amount includes the VAT), vat_only, and the validity dates — codes repeat across rate changes, so pick the one valid on the booking date. Read it before putting a VAT code on a booking. A company not registered for VAT has none to use.

list_vat_periods

ChatGPT
The company's VAT periods, oldest first: id (what update_vat_period takes), date_from, date_until (null = open-ended), is_vat_registered, vat_number, method ('effective' or 'net' = net tax rate, TDFN / Saldosteuersatz) and basis ('agreed' = VAT due on invoicing, 'collected' = on payment). Use it to answer whether and since when the company is VAT-registered and which method it uses on a date. A company with no period cannot book with VAT.

propose_bank_transaction_booking

ChatGPT
Prepares the booking that reconciles ONE imported bank transaction as a PROPOSAL the user applies with one click. The bank leg (amount, side and date) comes from the transaction itself — you name the cash account the transaction sits on and the counter account; money in debits the cash account and credits the counter account, money out the reverse. Transactions in a foreign currency need the Banking page. Attach the invoice or receipt the transaction pays with documents_ids — a bank booking cannot get a document afterwards through attach_document_to_booking. Before calling this tool, show the user the booking you are about to prepare (date, description, every leg with account and amount) and get their agreement; the tool then validates it and returns the resolved booking exactly as it would be saved — show that too. The result is the resolved booking (every leg with account, amount and the VAT the ledger derived), the proposal id, its expiry and a deep link: give the user the link — they apply the proposal in YESbusiness. You cannot apply it; apply_proposal needs a confirmation token only the app's card ever receives. Until applied it is a draft: not in the reports. Money rules: every amount is an INTEGER in CENTS of the company's HOME currency (read it with get_bookkeeping_guide or get_currency_settings first — never assume CHF); 12.50 is 1250. Account numbers, VAT codes and cost centers are the company's own — read them (list_accounts, list_vat_codes, get_cost_centers) before naming one. Reports are pre-summed: never add child rows yourself.

propose_booking

ChatGPT
Prepares a booking (a double-entry journal entry) as a PROPOSAL the user applies with one click: the legs are validated exactly as the booking panel validates a save, a draft is written with the connector as its source, and a proposal is filed. Before calling this tool, show the user the booking you are about to prepare (date, description, every leg with account and amount) and get their agreement; the tool then validates it and returns the resolved booking exactly as it would be saved — show that too. The result is the resolved booking (every leg with account, amount and the VAT the ledger derived), the proposal id, its expiry and a deep link: give the user the link — they apply the proposal in YESbusiness. You cannot apply it; apply_proposal needs a confirmation token only the app's card ever receives. Until applied it is a draft: not in the reports. For a bank transaction waiting to be booked use propose_bank_transaction_booking instead. Money rules: every amount is an INTEGER in CENTS of the company's HOME currency (read it with get_bookkeeping_guide or get_currency_settings first — never assume CHF); 12.50 is 1250. Account numbers, VAT codes and cost centers are the company's own — read them (list_accounts, list_vat_codes, get_cost_centers) before naming one. Reports are pre-summed: never add child rows yourself.

reject_proposal

ChatGPT
Rejects a pending proposal the user does not want: the proposal is closed and, for a booking proposal, its draft booking is deleted. Use it when the user says no to a booking you proposed; a proposal left alone expires on its own (24 hours).

revert_connector_writes

ChatGPT
Undoes bookings an assistant wrote (source connector:* or assistant): a draft is deleted, a complete booking is reversed by a new reversal booking dated today (or the last day of the latest open year). Only AI-written bookings in an open year, not linked to a customer invoice; a human's booking is never touched. The whole list is checked first and nothing happens unless every booking passes. On a connection that proposes (write:confirm) this returns the PLAN only — what would be deleted or reversed — and the deep link to the AI activity page, where the user runs the undo; on an autonomous connection (write:auto) the plan is run after the check. Show the user the plan before calling it.

update_account

ChatGPT
Changes an existing account of the chart: number, name, description, parent group, default VAT code, currency, the VAT-threshold flag, or the type (only while nothing is booked on it). Send only what changes; the rest is kept. account_type_id: 1 cash and bank, 2 other assets, 3 liabilities, 4 expenses, 5 revenue, 6 equity, 7 group (a header that takes no bookings). On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

update_accounting_year

ChatGPT
Changes an accounting year — its name, dates, status (closing a year = status 'blocked'; 'revised' once the audit is done; reopening = 'open') or booking-number format. All of name, date_from and date_until are sent (get_currency_settings lists the current values). An accounting year has a name, inclusive dates that must not overlap another year, a status (open takes bookings; blocked and revised are closed) and its booking-number format: prefix + the next number zero-padded to counter_size. On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

update_cost_center

ChatGPT
Renames a cost center (get_cost_centers gives the ids). On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

update_vat_code

ChatGPT
Changes an existing VAT code: name, code, rate, accounts, validity, report lines or whether it is offered. Send only what changes; the rest is kept. The regime fields (rate, kind, method) lock once bookings carry the code — a rate change is a NEW code valid from the new date, the old one ending the day before. rate is a FRACTION (0.081 for 8.1%, 0.026, 0.038, 0 for exempt). kind 'output' is VAT on sales (revenue), 'input' is deductible VAT on purchases. booked_in_account_number is the VAT account the tax is posted to (2200 VAT due for output, 1170 input VAT); entered_net says how a booking's amount is read: true = the amount EXCLUDES the VAT (it is added on top), false = the amount INCLUDES it (gross, the usual case); gross_in_account_number is required for a gross code. effective false marks a net tax rate (TDFN / Saldosteuersatz) code; vat_only a code that books the VAT alone (import VAT). On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

update_vat_period

ChatGPT
Changes an existing VAT period; send only what changes. A VAT period says, from a date, whether the company is VAT-registered, its VAT number, the method ('effective' or 'net' tax rate) and the basis ('agreed' = on invoicing, 'collected' = on payment). Periods must not overlap; date_until empty = open-ended. A change of method or basis is a new period starting on its first day. On a connection that proposes (write:confirm) nothing is written: the exact values and the deep link to the settings page come back for the user to confirm there; an autonomous connection (write:auto) writes directly.

upload_document

ChatGPT
Uploads a supporting document (a receipt, an invoice, a statement) into the company's Documents, and returns its document_id to attach to a booking (documents_ids on propose_booking / create_booking, or attach_document_to_booking). PDF, JPEG, PNG or WebP, 5 MB at most, sent as base64. The upload counts against the company's storage; only available when the connection was granted access to files. The file is stored as sent — never upload something the user did not give you for this purpose.

Capabilities

WritesInteractive

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/yesbusiness"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/yesbusiness)