claude-crew

Claude Crew provides an MCP server and a configuration CLI to build autonomous coding agents that leverage Claude Desktop. The server coordinates task reception, project information provisioning, and autonomous execution by running a pipeline that updates dependencies, refreshes the RAG index, and ingests repository data to feed the language model with context. Tools exposed by the MCP server enable you to run unit tests, lint, and perform type checks, while also facilitating project-specific actions like pulling latest changes and updating the environment. Use Claude Crew to deploy project-specific agents that can handle small, well-defined tasks automatically, which is especially useful for repetitive coding activities and maintaining consistent project pipelines.

To use the MCP server, install the package using the CLI (via npx), run the server to start listening for MCP tasks, and then configure Claude Desktop to point at the server's MCP configuration. The setup will generate a project-specific mcp.json with tool names that reflect your project, and the CLI will guide you through adding project details and instructions. Once running, the server will orchestrate task reception, project information provisioning (prepare), and autonomous execution with linting, testing, and necessary code changes, while transmitting optimized context to the LLM throughout the workflow.

Prerequisites:

• Node.js v20 or newer installed on your machine
• npm or corepack available
• Claude Desktop installed (optional but recommended for context integration)

Installation steps:

1. Ensure Node.js is installed
   • Check: node -v
   • If not installed, download from https://nodejs.org/ and install v20+
2. Install and run the MCP server via npx
   • Ensure you have network access to fetch the package
   • Run the setup command to initialize configuration (interactive):

$ npx claude-crew@latest setup

3. Start the MCP server (as defined by the package's server entrypoint)

$ npx claude-crew@latest server

4. Connect Claude Desktop to the MCP server by adding the generated .claude-crew/mcp.json configuration to Claude Desktop as described in the Quick Start of the README.

Note: The configuration file (.claude-crew/mcp.json) will be created during setup and includes project-specific tool definitions.

Tips and notes:

• Claude Crew relies on a Claude Desktop environment and optional OpenAI embeddings for improved context understanding. Ensure you have an OpenAI API key if you plan to use embeddings.
• The server creates project-specific configurations to avoid cross-project conflicts; duplicate projects can be managed by running separate setups.
• If the MCP server fails to start, verify that your claude_desktop_config.json matches the latest format generated by .claude-crew/mcp.json and refer to config.example.json for the correct structure.
• Typical commands exposed via the CLI include setup, clean (to reset the database), and setup-db (to initialize the embeddings database).