Groundcheck

Please install the `groundcheck` MCP server into my current AI client (that's you).

Required prerequisites (do these first if not already done):
- **Install Groundcheck dependencies and build the server** — Set up both the Python engine and Node server from the repository, then compile the TypeScript MCP server. Run: `make install && npm --prefix server run build`
- **Provide an LLM provider API key for stance classification** — Set a router provider key such as Groq to enable grounded verdicts. Without a provider key, retrieval works but verdicts degrade to 'unverified'.

Canonical MCP server config (stdio transport):
- command: `node`
- args: ["<path-to-groundcheck-repo>/server/dist/server.js"]
- required environment variables:
  - `GROQ_API_KEY`: API key for Groq or another supported router provider. Required in practice for Groundcheck to return supported/refuted verdicts rather than only unverified results. (example: `<your-groq-api-key>`)
- optional environment variables:
  - `GROUNDCHECK_ENGINE_URL`: HTTP URL where the MCP server reaches the Python engine. Defaults to the local engine at 127.0.0.1:8723. (example: `<your-groundcheck-engine-url>`)
  - `GROUNDCHECK_NO_SPAWN`: Set to disable auto-spawning the Python engine from the MCP server. (example: `<disable-auto-spawn-flag>`)
  - `GROUNDCHECK_ENGINE_DIR`: Path to the engine directory used when the MCP server auto-spawns the Python engine. (example: `<path-to-engine-dir>`)
  - `GROUNDCHECK_PYTHON`: Python interpreter used by the MCP server when auto-spawning the engine. (example: `<path-to-python>`)
  - `GROUNDCHECK_REPO_URL`: Repository URL used in attribution footer/badge generation. (example: `<your-repo-url>`)
  - `GROUNDCHECK_SEARCH_BACKEND`: Engine setting to control retrieval backend; set to stub to disable real retrieval. (example: `<search-backend>`)
  - `GROUNDCHECK_SEARCH_URL`: Custom JSON search endpoint for retrieval, returning results in the expected schema. (example: `<your-search-endpoint-url>`)
  - `GROUNDCHECK_SEARCH_KEY`: Bearer token for the custom search endpoint. (example: `<your-search-api-key>`)
  - `GROUNDCHECK_ROUTER_PATH`: Path to the free-llm-router Python package used by the engine. (example: `<path-to-free-llm-router>`)
  - `GROUNDCHECK_ENGINE_HOST`: Bind host for the Python engine. (example: `<engine-host>`)
  - `GROUNDCHECK_ENGINE_PORT`: Bind port for the Python engine. (example: `<engine-port>`)
  - `GROUNDCHECK_ENSEMBLE`: Enable multi-provider stance panel; 0 means single-router mode. (example: `<ensemble-enabled-flag>`)
  - `GROUNDCHECK_ENSEMBLE_MAX`: Maximum concurrent panelists per claim for ensemble stance classification. (example: `<max-panelists>`)
  - `GROUNDCHECK_CALIBRATION`: Path to the conformal calibration artifact JSON file. (example: `<path-to-calibration-json>`)
  - `GROUNDCHECK_X402_PAY_TO`: Enables machine-payable hosting features on the engine when set. (example: `<your-x402-pay-to-address>`)

Note: Preferred transport for local installation is stdio via the Node MCP server. The server auto-starts the Python engine if needed. The README also mentions a future npm package via 'npx -y groundcheck-mcp', but it is not yet published, so the repository build path is the usable install path. A remote MCP endpoint is also documented at https://groundcheck.seiche.info/mcp, but the canonical local install shape in this README is stdio.

Add this MCP server to my current client's config in the correct format for you. If you need secrets or credentials I haven't provided, ASK me — do not invent values or leave raw placeholders. After adding it, tell me how to verify the server is connected.