Model Context Protocol
Connect Claude or ChatGPT to your books over MCP. It reconciles your accounts and reproduces the Schedule E you filed.
The demo books are connectable read-only. Your own books are by invite. Tool calls cost nothing on top of the AI you already use.
https://www.oberlin24.com/api/v1/mcp
Not a chatbot bolted onto a bank feed. Your AI calls the same tax engine the app runs, so the numbers tie to your dashboard and your filing.
Diff your live books against the Schedule E you actually filed, per property and per line. The trust check a generic ledger cannot do.
reproduce_filed_returnEvery transaction on the right IRS line, per property, ready to file.
get_schedule_eMACRS and straight-line schedules feeding line 18, carried from your filed run rate.
depreciation_statusEach account tied to the bank, pending settling shown separately, never a phantom gap.
reconciliation_statusThe BAR test and the 2,500 de-minimis safe harbor on a repair, decided for you.
assess_capitalizationGain, depreciation recapture, suspended-loss release, and the 1031 comparison, per property.
decision_analysisYour Section 199A deduction and a quarterly estimate off the live numbers.
qbi_estimateExpenses hiding in the wrong place, surfaced with the dollars they add back.
find_missing_deductionsAdd the server once with the URL above. Reads run on their own; every change comes back as a preview you confirm before anything is written.
https://www.oberlin24.com/api/v1/mcp
https://www.oberlin24.com/api/v1/mcp
--scope user makes the server available in every project; without it the default scope is local, so the server only exists in the folder where you ran the command./mcp, pick oberlin24, and choose Authenticate. Sign in on the Oberlin24 page that opens.claude mcp add --transport http --scope user \
oberlin24 https://www.oberlin24.com/api/v1/mcp
[mcp_servers] in ~/.codex/config.toml.codex mcp add oberlin24 \
--url https://www.oberlin24.com/api/v1/mcp
codex mcp login oberlin24
Sign in to the demo first to try it read-only against sample books. Connecting your own books is by invite.
A general ledger tags a transaction and hands the hard part to your CPA. This one knows real-estate tax, so your AI can answer the questions the CPA usually does.
Every tool call your assistant makes runs as plain code on our side. No model spins up, no metered AI spend, no per-call fees. The thinking happens in the AI subscription you already pay for, so a full bookkeeping session adds zero to your bill.
Your AI reads freely. A change is never made silently. It comes back as a preview with an approval token, and nothing is written until you confirm it. Access is scoped, keys are yours to revoke, and the demo is read-only.
Try it on the live demo, or ask for an invite to connect your own.