Arranger MCP Server
The Arranger MCP server sits on top of Arranger (deployed with the search portal above) and exposes your indexed data to MCP-compatible AI clients, so researchers can query the portal in plain language. Deploy it only if you want to offer AI-assisted, conversational data discovery on top of your search portal.
For the workflow this enables, the researcher experience, and a locally deployable demo, see the AI-Assisted Data Discovery guide.
Running the MCP server
-
Run the MCP server: Point the container at the Arranger you deployed with the search portal.
docker pull ghcr.io/overture-stack/arranger-mcp-server:b4a414cedocker run -d \--name arranger-mcp \-e ARRANGER_BASE_URL=http://<your-arranger-host>:5050 \-e ARRANGER_CATALOGUES=<your-catalogues> \-p 3100:3100 \ghcr.io/overture-stack/arranger-mcp-server:b4a414ceReplace
<your-arranger-host>with your Arranger's hostname or IP. If Arranger runs in Docker on the same host, usehost.docker.internal(macOS/Windows) or the Arranger container's service name on a shared network. SetARRANGER_CATALOGUESto the catalogues your Arranger actually serves; the server exits on startup if that list is missing or empty.Variable Default Required Description ARRANGER_BASE_URLhttp://localhost:5050Required Base URL of the Arranger server ARRANGER_CATALOGUESserverRequired Comma-separated list of Arranger catalogues to expose ARRANGER_REQUEST_TIMEOUT_MS10000Optional HTTP timeout in milliseconds for requests to Arranger MCP_HOST0.0.0.0Optional Host the MCP server binds to MCP_PORT3100Optional Port the MCP server listens on MCP_PATH/mcpOptional Endpoint path for the Streamable HTTP transport LOG_LEVELinfoOptional Pino log level ( trace,debug,info,warn,error,fatal) -
Confirm the server is up: Check that the container is running and connected to Arranger.
docker ps | grep arranger-mcpdocker logs arranger-mcp # should show it connected to Arranger and listeningThe server is now reachable at
http://<your-arranger-host>:3100/mcpover the Streamable HTTP transport, ready for an MCP-compatible client to connect.
At LOG_LEVEL debug or trace, the server may write SQON query objects to its logs. Those queries are generated from researcher input and can reveal which datasets, filters, and value ranges a researcher is investigating. Treat log files as potentially sensitive data: keep LOG_LEVEL at info or higher in production unless you are actively debugging, and apply the same access controls and retention policies you use for your primary portal's query logs.
For the server's internals, its full tool inventory, and development and testing workflows, see the canonical Arranger MCP Server documentation at apps/mcp-server and Arranger's MCP server reference.
This guide was verified against the Arranger MCP server image arranger-mcp-server:b4a414ce, built from Arranger v3.0.0-beta.32+8e84b4b, on 2026-07-28. Newer releases may differ.