repobase
Index and search your Git repositories with AI. Includes a terminal UI and MCP server for AI tool integration.
claude mcp add --transport stdio fernandoabolafio-repobase repobase-mcp
How to use
Repobase is a tool to index and search your Git repositories with AI, featuring a terminal UI and an MCP server for integration with AI tools like Cursor and Claude. The MCP server exposes a set of useful commands that let you interact with your repositories from within AI prompts or automation: list_repos to enumerate all indexed repositories, search to query across repos with keyword, semantic, or hybrid modes, list_files to inspect repository contents, glob_files to find files matching a glob pattern, read_file to fetch file contents, and grep to search file contents using regular expressions. To enable AI-driven workflows, configure the MCP in your AI tool's MCP config (for example in ~/.cursor/mcp.json or Claude config) to point to the repobase-mcp executable, then call the exposed tools as needed within your prompts or tooling integration.
Once configured, you can use the MCP server to programmatically list repositories, run searches across your codebase, and inspect files or snippets directly from your AI-assisted workflows. This makes it easier to locate relevant code, review project structure, or extract snippets for documentation or debugging without leaving your AI tool or environment.
How to install
Prerequisites:
- Bun runtime (recommended) as described in the Repobase docs
- Node.js-compatible tooling is optional if you use Bun; the project ships an MCP server binary wrapper.
Installation steps:
- Install Bun (if not already installed): curl https://bun.sh/install | bash
- Install Repobase globally (or install dependencies and use the provided MCP module):
npm install -g repobase
Alternatively, using Bun:
bun install -g repobase - Run the MCP server in development or production mode as needed: repobase-mcp
- Verify the MCP server is reachable by your AI tool configuration (e.g., at the default local endpoint the host expects for MCP servers).
Upgrade path and local development follow the project’s standard Bun-based workflow, including building if you’re contributing to the MCP server codebase and running tests with bun run test.
Additional notes
Tips and notes:
- The MCP server is named repobase and exposes tools for repository indexing and search. Ensure your AI tool's MCP config points to the correct command (repobase-mcp).
- Environment variables are optional for basic usage; you can add variables in the mcp_config under env if your setup requires API keys or custom configuration.
- The repository ships a terminal UI (for local interaction) and a separate MCP server module intended for AI tooling, so you can both interact manually and automate queries from AI prompts.
- If you encounter issues connecting from Cursor or Claude, verify network access to the MCP server, correct command path, and that the MCP server has permissions to access the repositories you want to index.
- The npm package name is repobase, which corresponds to the Node ecosystem distribution used for the MCP integration.
Related MCP Servers
gemini-cli
An open-source AI agent that brings the power of Gemini directly into your terminal.
mcp -azure-devops
An MCP server for Azure DevOps
jigsawstack
Model Context Protocol Server that allows AI models to interact with JigsawStack models!
mcp-image-extractor
MCP server which allow LLM in agent mode to analyze image whenever it needs
mcp-rfc
An MCP server for fetching RFC related information.
GameMaker
GameMaker MCP server for Cursor - Build GM projects with AI