Spins up Aider as a detached background process so your LLM can kick off code editing tasks and check back later on progress. The server creates tracking files including process details, chat history, and the original prompt, then exposes tools to monitor git status, read chat logs, and verify if Aider is still running. Requires whitelisting specific directories via glob patterns for safety. Also includes basic git tools for branch switching, stashing, and reverting commits, plus file reading capabilities. The architecture lets you build agentic workflows where one LLM delegates coding work to Aider instances and polls their markdown chat history files to see what got done. Needs Node, Aider, and git installed locally, and you configure your LLM backend through environment variables pointing to OpenAI compatible endpoints.
claude mcp add --transport stdio railkill-mcp-aider-executor -- npx -y mcp-aider-executor