Skip to main content
Your agent can grep, glob, and read your local codebase, but it cannot do the same thing across the open-source code your app depends on. That is where it breaks: niche integrations, undocumented APIs, version-specific behavior, and dependency internals. It starts guessing, retries small changes, then pushes you back into search, docs, forums, and trial-and-error. GitHits gives AI coding agents context from public open-source software. Agents get version-aware access to indexed code, package docs, package metadata, vulnerabilities, changelogs, dependency graphs, upgrade evidence, and real implementation examples. GitHits builds its index on demand, in about 10 seconds for an average repo, so agents can retrieve and navigate the actual code running in your stack. It gives them:
  1. Code across packages and repos: search, grep, file listing, exact line reads without cloning, and implementation examples.
  2. Documentation across hosted docs and repo-backed docs.
  3. Package Intelligence for dependencies, vulnerabilities, changelogs, dependency graphs, and upgrade changes.

The problem GitHits solves

AI coding agents are strongest when they can inspect the system in front of them. Locally, they can search your files, read exact lines, and follow definitions. Across dependencies, they usually lose that ability. Package docs may be incomplete. APIs change by version. Internals are rarely documented. Many useful patterns live in real implementations, issues, discussions, and pull requests rather than in official guides. GitHits bridges that gap with searchable context from public open-source software. Source navigation, documentation, and package inspection support version-specific research. Within Code, get_example finds prior art and current implementation patterns when you do not need a version-pinned source inspection.

How it works

GitHits runs through a hybrid CLI that configures the verified MCP transport for your AI coding tool. Most direct installs use local stdio. Cursor and plugin or extension installs use the hosted remote MCP. Once connected, your agent can call GitHits tools when it needs context from indexed public open-source code, package docs, package metadata, or examples. If your harness is conservative about tool use, you can trigger GitHits manually.
init detects supported coding tools installed on your machine and configures each one with the GitHits MCP server. Authentication is handled automatically for common local setup, so you usually do not need to think about it.

MCP tools

GitHits organizes its evidence tools into three categories: Code, Documentation, and Package Intelligence. Session guidance helps agents choose the right tool.

Code

Search, grep, list files, and read exact line ranges across indexed packages and repositories without cloning. Use get_example for prior art and implementation patterns from repositories, issues, discussions, and pull requests.

Documentation

Search, browse, and read hosted docs and repo-backed docs for indexed packages.

Package Intelligence

Inspect dependencies, vulnerabilities, changelogs, and upgrade changes.

Supporting tools

Plain MCP clients use quick_start to load the current routing and safety guide. The packaged githits-mcp skill already includes the stable guide.

Automatic install support

npx githits@latest init supports automatic install for Claude Code, Cursor, Windsurf, Claude Desktop, Codex CLI, Pi, VS Code / Copilot, Cline, Gemini CLI, Google Antigravity, OpenCode, Hermes Agent, Zed, Junie, Qwen Code, Kiro, Kilo Code, Factory Droid, and Amazon Q CLI. For any MCP-capable client that supports remote servers, use Manual remote MCP setup.

Next steps

Connect GitHits with the quickstart, or use CLI-backed Skills. Then explore the three tool categories:

Code

Search, grep, list files, read source, and find implementation examples.

Documentation

Search, browse, and read hosted and repo-backed package docs.

Package Intelligence

Review metadata, vulnerabilities, dependencies, changelogs, and upgrade evidence.