Mercury is one of the go-to banks for AI-native businesses, with a full treasury offering: operating checking, savings accounts, treasury returns, and cards for team and agent spend. Every account generates rich, well-structured data with counterparties, memos, and timestamps. This is all the raw material for a live picture of the business’s cash.
The skill is in interpreting it correctly. Connect Mercury to AgenticBooks and that raw data becomes useful management information: internal movements matched, real activity booked, yield recognized properly, and cash, burn, and runway instantly available to founders and agents.
Interpreting Mercury activity correctly
A bank feed records cash movements, and accounting records business activity. The difference is a core rule of bookkeeping. Some movements through Mercury are the business: customer settlements, contractor payments, API bills charged to cards, etc. Others are the company’s own money changing pockets: the weekly sweep into the Treasury account, interbank movements, currency conversions, and card payments.
Interpreted properly, each Mercury account tells its part of the story:
- Operating checking is the business happening. Settlements in, vendors and contractors out. These must be booked to their correct revenue and expense accounts in the period they occur.
- Internal movements are transfers, not activity. A sweep into the Treasury account is one movement of the company’s own money. Booked as a single inter-bank transfer, it touches neither revenue nor expenses, so the burn rate reflects actual spending and excludes internal movements.
- Treasury yield is interest income. An income line, visible and distinct from product revenue, keeping the top line segregated from the investment yield.
- Card spend is the expense; the repayment is a transfer. The API charges, tooling, and infrastructure on the cards are the actual costs, categorized individually. A lump-sum repayment from checking is just money moving.
Get this interpretation right and the numbers become quality management information: accurate burn rate, runway on live balances, and spend analysis reflecting actual business consumption. AgenticBooks does this interpretation automatically, transaction by transaction, as activity arrives.
Automatic reconciliation: every movement matched
Interpretation is one half and reconciliation is the other. Every internal movement produces two records and the ledger needs to tie the two legs together as a single transfer.
AgenticBooks reconciles them continuously. When a sweep, savings transfer, or repayment moves between Mercury accounts (or between Mercury and any other connected bank), the two legs are paired and proposed as a suggested transfer. Confirming books the pair as one movement; rejecting returns both events to the normal flow, because a customer payment that coincidentally mirrors a vendor bill is not a transfer. Both outcomes are logged actions, so the queue is safe to delegate to an agent, and the ledger, the bank, and the true cash position agree with each other continuously.
This continuous agentic accounting is transformative. The classic quarter-end exercise (untangling months of sweeps and repayments to find the real burn number before an investor update) becomes redundant. Timing differences, cross-bank movements, and cross-currency transfers are all handled in the matching; anything ambiguous is clarified, with every resolution logged in the audit trail.
Live reporting: cash intelligence in any form
With AgenticBooks’ MCP connection, all of the fully interpreted and reconciled Mercury data is available for any data questions, analysis, and reporting.
Connect ChatGPT, Claude, or any LLM and the data is fully available for queries: “Current runway at this month’s burn?” “Real burn rate over the last two quarters, and the trend?” “Model API spend on the cards versus revenue, by month?” “How much is the Treasury yield offsetting monthly costs?” “Post the Monday cash position to Slack.” “Produce a board pack for this quarter.”
Skip the build: no manual Mercury API integration required
AI-native teams can use the Mercury API to build their custom data flows. However, agentic accounting is much deeper than purely obtaining the raw data: transfer-matching logic with timing tolerances, categorization rules that survive edge cases, correct treatment of Treasury yield, cross-currency movements, audit trails, and close semantics. That’s specific finance and accounting domain knowledge.
AgenticBooks is that entire layer combining the data and agentic accounting: connect Mercury, connect any other banks, and the interpretation, matching, and reconciliation run correctly from day one. The engineering attention stays on the product and the AI tools get the finished financial data over MCP.
FAQ
Does AgenticBooks cover all Mercury account types?
Yes. Checking, savings, and Treasury are all included in the unified balance view, alongside any other connected banks.
How is Mercury Treasury yield booked?
As interest income on its own line, never mixed into product revenue. Principal sweeps in and out of Treasury are booked as internal transfers.
Does transfer matching work between Mercury and other banks?
Yes. Matching pairs movements across all connected institutions, including cross-bank and cross-currency transfers.
What happens to a suggested transfer that is actually a coincidence?
It gets rejected, returning both events to the normal review flow. Confirm and reject are both explicit, logged actions on the audit trail.
Do I need to write any code against the Mercury API?
No. Connect Mercury and any other banks to AgenticBooks and the full interpretation, matching, and reconciliation layer runs without an integration project.
AgenticBooks is the accounting layer for AI-native businesses — a ledger your agents can operate over MCP, with unified multi-bank balances, automatic transfer matching, and one-command month-end close.