J.Jupid Docs
MCP for Jupid users

Jupid MCP

Connect external AI agents to Jupid financial data through hosted MCP.

Jupid MCP is a hosted Model Context Protocol endpoint for external AI agents. It lets MCP-capable clients answer questions using the same user-scoped Jupid data available in the app.

Use it when you want an assistant such as Claude, Codex, Cursor, ChatGPT, VS Code with GitHub Copilot, or Windsurf to inspect Jupid financial context without exporting data manually.

How it works

The hosted MCP server URL is stable for all users:

https://backend.jupid.com/mcp

The endpoint uses Streamable HTTP (JSON responses) and OAuth 2.1 with PKCE. When a client first connects, Jupid opens a browser sign-in flow and authorizes the client for the signed-in user — no API keys. Access tokens expire after one hour and refresh automatically for up to 30 days.

The MCP server is user-scoped. A connected client can only access data that the authorized Jupid user can access in the app.

What agents can answer

  • Accounts, balances, integrations, and sync status.
  • Transactions and counterparties — vendors, clients, investors, contractors — by date, amount, category, or review state.
  • Cash-flow and Schedule C P&L reports by period and category tree.
  • Open questions that need your context before Jupid can keep processing.

See MCP tools for the full catalog with inputs and outputs.

Safety boundaries

Tools return structured summaries, not raw database rows. Account numbers are masked, search results are bounded, and every result links back to the app for review.

Twelve of the thirteen tools are read-only. The only write-capable tool is answerOutstandingIssue: it saves an explicit answer to one existing open issue, and the answer goes through Jupid's normal processing. It does not recategorize transactions. There is no recategorization, category editing, file upload, account sync trigger, billing change, or row editing through MCP today — see Not available yet.

Next steps

On this page