add-third-party-tool

To expose a CLI tool as an MCP endpoint, you wrap it by defining its install commands, entrypoint, and operation schemas in a .mica/tools.json manifest, allowing the agent to invoke it reliably with structured inputs.

No, you do not need custom server code to integrate a command-line tool. This approach uses a YAML-described workflow to map the tool's entrypoint and operations directly into a project-scoped MCP tool manifest.

You define structured inputs and outputs for a wrapped CLI tool by declaring per-operation schemas that specify argv templating or stdin/stdout IO modes, ensuring the agent sends correct arguments and receives predictable results.

Yes, you can automate the installation lifecycle by declaring install steps in the manifest so the wrapper handles first-time setup and caching, or setting it as a no-op if the tool is already on PATH.

Limitations of wrapping CLI tools include requiring bounded timeouts for operations and dependency on a YAML-described workflow mapping to function, meaning tools needing persistent state or interactive UI cannot be wrapped reliably.

The best way to add an external tool not on the agent surface is to wrap it into a project-scoped MCP tool endpoint by updating .mica/tools.json with the tool’s install command, entrypoint, and at least one operation schema.