uol-mcp-http

HTTP MCP server for the UOL Accounting REST API (Streamable HTTP for Railway and Cursor).

Reads run immediately. POST / PATCH / PUT / DELETE go through uol_request and need confirm=true.

Env

• UOL_API_BASE_URL — tenant API root, e.g. https://YOUR-TENANT.ucetnictvi.uol.cz/api (with or without /v1)

• UOL_API_EMAIL — UOL login email (Basic Auth username)

• UOL_API_TOKEN — token from UOL Settings → /api_tokens

• MCP_AUTH_TOKEN — Bearer token required by MCP clients

• PORT — listen port (Railway sets this)

Demo (OpenAPI): https://test.demo.uol.cz/api + the demo email/token from https://api.uol.cz/

Related MCP server: accounting-mcp

Endpoints

• GET /health — healthcheck

• POST /mcp — MCP Streamable HTTP (auth required)

Cursor

Do not commit real tokens. Add this locally in .cursor/mcp.json:

{
  "mcpServers": {
    "uol": {
      "url": "https://uol-mcp-production.up.railway.app/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_MCP_AUTH_TOKEN"
      }
    }
  }
}

YOUR_MCP_AUTH_TOKEN is MCP_AUTH_TOKEN in local .deploy-secrets (gitignored) and on the Railway service.

This instance currently talks to the UOL demo tenant (https://test.demo.uol.cz/api). For live Gramodesky accounting, change Railway vars UOL_API_BASE_URL / UOL_API_EMAIL / UOL_API_TOKEN to the production tenant (same values as shop UOL_API_*).

GramBot (later)

When the shop worktree is clean, attach HTTP MCP with MCP_UOL_URL + MCP_UOL_TOKEN (same pattern as Daktela / Fio).

Rate limit

UOL returns HTTP 429 for 30 seconds when exceeded: 30 requests / 10 seconds (receivables: 10 / 10 seconds).