Guru MoneyGuru MoneyFinance workspace
Recommended access model

The current convenient model is a personal MCP key linked to a personal Guru Money access token. This lets the AI client act on behalf of a specific user instead of sharing one broad token.

Supported clients

ChatGPT, Claude, Claude Code and other clients that can speak MCP over stdio or HTTP.

Transport modes

Both local stdio mode and production-like HTTP mode behind Traefik are supported.

Technical docs

Detailed technical documentation is available at https://docs.guru-money.com and in the repository MCP materials.

How authentication works

  • In per-user mode the server reads `MCP_USER_TOKENS` and maps an MCP key to a Guru Money access token.
  • Clients can send `Authorization: Bearer <USER_MCP_TOKEN>` or use a compatible query token fallback when needed.
  • A legacy shared-token mode still exists, but the preferred approach is per-user MCP access.

What the MCP layer can do

  • Read session, workspaces, accounts, categories, limits and transactions.
  • Work with bank connections, including the Monobank flow.
  • Trigger selected finance actions through the API-backed tool layer.

Why this design is useful

The MCP service stays a thin layer on top of `api`, which reduces duplicated logic and avoids direct database exposure.

For production use, the project already includes health checks, Traefik routing and deployment guidance for the HTTP MCP service.