Atomic Chat — локальные модели через MCP

★ 7.5 · ai-tooling

add-atomic-chat-tool is a Claude Code skill that adds a stdio-based MCP server so the container agent can call local models served by the Atomic Chat desktop app via its OpenAI-compatible API at `http://127.0.0.1:1337/v1`. Claude acts as the orchestrator and can offload work to locally running models such as Llama or Qwen without sending data to external services. Two tools are exposed: `atomic_chat_list_models` (GET /v1/models) to enumerate available models, and `atomic_chat_generate` (POST /v1/chat/completions) to send a prompt and retrieve a response. Installation copies the MCP server source and tests into the agent-runner tree, registers the server in `index.ts`, and wires host environment variable forwarding through `container-runner.ts`. Atomic Chat itself is a fork of Jan and handles model management through its own desktop UI.