Skip to main content
GitHits works by running a local MCP server that your AI coding tool starts on demand. GitHits gives AI coding agents version-aware access to indexed public open-source code, package docs, package metadata, vulnerabilities, changelogs, dependency graphs, upgrade evidence, and real implementation examples. Beta: tool names, setup flows, and result formats may change as GitHits evolves.
1

Check prerequisites

GitHits requires Node.js 20 or later. To check your version, run:
If you need to install or upgrade Node.js, visit nodejs.org.
2

Run the hybrid CLI

From your terminal, run:
This command:
  1. Detects supported AI coding tools installed on your machine
  2. Configures each detected tool with the GitHits MCP server
  3. Handles authentication automatically for common local setup
init supports automatic install for Claude Code, Cursor, Windsurf, VS Code / Copilot, Cline, Claude Desktop, Codex CLI, Pi, Gemini CLI, Google Antigravity, and OpenCode.
If you use any MCP-capable client that supports remote servers, connect it to GitHits manually. See Manual remote MCP setup.
3

Use your agent

Open your AI coding tool and start a coding task as normal. Many agents call GitHits tools on their own when they need indexed source, docs, package metadata, or implementation examples.Some harnesses are more conservative. If the agent starts guessing or does not call GitHits, ask for it directly:
For example, if you ask your agent to debug Express middleware behavior, it may search the indexed package and read the relevant source:
The agent receives indexed source context and uses it to explain the dependency behavior before changing your code.

​
What’s next