codemcp
@ezyang
About codemcp
Coding assistant MCP for Claude Desktop
Overview
What is codemcp?
codemcp is a Model Context Protocol (MCP) server that turns Claude into a pair programming assistant. It runs on your computer and is intended for Claude Pro subscribers ($20/month), working with Claude Desktop or the web version via a browser extension. Developers can ask Claude to implement features, fix bugs, and refactor codebases directly on their machine.
How to use codemcp?
Install uv and git, then install the claude-mcp browser extension. Run uvx --from git+https://github.com/ezyang/codemcp@prod codemcp serve to start the MCP server and configure the extension with the SSE URL (e.g., http://127.0.0.1:8000/sse). In Claude, create a project with instructions like Initialize codemcp with $PROJECT_DIR, and create a codemcp.toml file in your Git repository to define allowed commands (e.g., format, test). Then chat with Claude to make changes.
Key features of codemcp
- Auto‑accept changes by default with automatic Git commits for rollback
- Pre‑declared commands in codemcp.toml restrict shell access securely
- Works with any IDE – Claude edits files, you review in your favorite editor
- Zero per‑action cost thanks to Claude Pro’s $20/month subscription
- Open source – modify or extend as needed
- Format command runs automatically after every file edit
Use cases of codemcp
- Implement a new feature by describing the requirement to Claude
- Fix
Basic information
More Coding Agents MCP clients
Aider MCP Client: Supercharge Your Coding with AI-Powered Documentation
alvinveroyA Python tool to query Context7 MCP for library docs. Integrates with Aider for AI coding. Install via pip install mcp_query and fetch docs easily.
UtilityBelt
richardanayaTalk to MCP servers from aider

Mitii AI Agent
Mitii-devYour local-first AI coding agent for complex work. Read files, write code, run commands — all with your approval.
Install MCP CLI
dhravyaA simple CLI to install MCP servers into any client
Overreach
Naveja00Overreach catches AI-agent scope creep. You give it the prompt you gave your coding agent and the diff it produced; it flags every part of the diff the prompt didn't authorize — unauthorized dependencies, env vars, endpoints, cron jobs, files, and features. Every finding is deter
Comments