Skip to main content
Ten tools cover everything an assistant does with transactions. Read tools take a view; write tools take an action; process_receipt takes a mode. Full parameters and return shapes for every tool are on the tools reference — this page is the map.
Every request is scoped to your own account. Transaction-reading tools return everything in your ledger: manual entries, CSV imports, API-created rows and transactions synced from bank accounts you have connected.

Reading transactions

Use filter for anything expressible as a field match (startDate, endDate, merchant, q, categoryId, isTaxDeductible, minAmount, maxAmount, amountEquals, page, pageSize). Use intelligence with keywords[], transactionTypes[] and categories[] when the category may not be set yet. analyze_spending and get_tax_summary accept a period preset (this_month, last_month, this_quarter, last_quarter, this_fy, last_fy, this_year, last_year, ytd, all_time) so you don’t compute date bounds yourself.
Returns follow one envelope: { success: true, data: { ... } }. query_transactions returns data.transactions, data.totalCount and a pagination block; get_tax_summary returns totalCount, totalAmount, byCategory and dateRange, plus confidence and GST/VAT blocks when you pass includeConfidenceAnalysis or includeGSTBreakdown.

Adding transactions

Both run your custom rules on the new record and report which ones matched. Both are retry-safe: the same amount, date and description within 10 minutes returns the existing record with alreadyExisted: true instead of a duplicate.

Editing transactions

categorize works by merchant without a lookup pass first — “categorise everything from Officeworks as Office Expenses” is one call. analyze_transactions suggests but never changes your own classification; feed what the user accepts back through modify_transactions with action: "apply_analysis".
smart_categorize is registered but not yet served over MCP — it returns a not-available response. Use action: "categorize" per transaction in the meantime.

Receipts

Both need the receipt_capture feature (Pro, Elite+). smart is the recommended path; use analyze_only then commit when the user should confirm the extracted fields in between.
The response carries uploadUrl, the exact headers to PUT with, expiresAt, and a ready-made next call — process_receipt with mode: "smart" and the assetUrl.

Good to know

  • Amounts. Negative is money out, positive is money in. create_expense takes the positive amount of the expense.
  • Confirmation. Clients other than Meta Muse (which asks the user itself) get a needsConfirmation preview on the first write call and repeat it with confirmToken to execute. See Authentication.
  • Rate limits. analyze_spending and create_transaction allow 30 calls a minute, modify_transactions and analyze_transactions 20, process_receipt and get_receipt_upload_url 10. Everything else has only the global limit.
  • Older names. Prompts or scripts written against the pre-consolidation tool names will fail with an unknown-tool error — the rename table says where each one went.

Next steps

Bill tools

Recurring bills, payments and what’s coming up

Full tools reference

Every parameter and return shape