add-atomic-chat-tool

Copy the MCP server source into container/agent-runner/src, register an atomic_chat entry in the mcpServers object in index.ts, forward ATOMIC_CHAT_HOST env vars in container-runner.ts, then rebuild the container. The agent then gains atomic_chat_list_models and atomic_chat_generate tools.

The MCP server defaults to http://host.docker.internal:1337 and falls back to localhost if that fails. You can override the endpoint by setting ATOMIC_CHAT_HOST in your .env file, which is forwarded into the container as a Docker -e argument.

No, Atomic Chat does not require authentication when running locally, so ATOMIC_CHAT_API_KEY can be left unset. Only set it if you have placed Atomic Chat behind a reverse proxy that enforces authentication.

This means the MCP server was not copied, not registered in the mcpServers object in index.ts, or the container was not rebuilt. Check that atomic-chat-mcp-stdio.ts exists in the agent-runner tree and run ./container/build.sh again.

The model ID passed to atomic_chat_generate must exactly match one of the IDs returned by atomic_chat_list_models. Ask the agent to list models first, then pick one from that list before generating.

Atomic Chat lazy-loads models into memory on first use, so the initial call takes longer while the model warms up. Subsequent calls against the same model are fast.