I’m Olli-Pekka, co-founder of GitHits (and the creator of opencv-python package).
GitHits is a CLI and local MCP server that gives coding agents access to the open-source code their projects actually depend on.
Coding agents already have access to your local codebase. The missing context is the open-source code behind your dependencies.
When an agent hits a framework edge case, an undocumented integration, or version-specific behavior, the answer often lives in dependency source code, tests, changelogs, issues, or discussions rather than in the application code itself. Without access to that information, agents spend more time exploring dead ends, retrying approaches, and consuming tokens before converging on a solution.
During the private beta, GitHits focused on generating examples. You could describe a problem in natural language, and it would search open-source repositories, issues, discussions, and package metadata for similar implementations, then generate an example grounded in real code.
The public beta expands GitHits beyond example generation. Agents can now move beyond finding examples and inspect the dependencies they use directly: searching source code, accessing documentation, inspecting dependency graphs, vulnerabilities and changelogs, reviewing package upgrades, and retrieving information for the specific package versions used by a project.
No repo cloning is needed. We have built a version-aware index of open-source code and documentation. Popular projects are pre-indexed across ecosystems. If a package or repository is missing, we index it on demand, typically in 10-20 seconds for an average-sized project.
The CLI and local MCP server currently expose the following tools:
- get_example for implementation examples from open source
- search, code_grep, code_files, code_read for code navigation
- docs_list, docs_read for documentation access
- pkg_info, pkg_deps, pkg_vulns, pkg_changelog, pkg_upgrade_review for package inspection
Package navigation currently supports npm, PyPI, crates.io, Maven, NuGet, Go modules, RubyGems, Packagist, Hex, Zig, vcpkg and Swift.
Code navigation currently supports C++, C#, Dart, Elixir, Erlang, Go, Java, JavaScript, Kotlin, Lua, PHP, Python, R, Ruby, Rust, Scala, Swift, TypeScript and Zig. We also have more limited support for Bash, CSS, HTML, Markdown, Proto, SCSS, Svelte, Vue. We are constantly adding support for more ecosystems, and advanced navigation tools over our indexed code graph are also on the roadmap.
We benchmarked Codex on a real integration task with and without GitHits. With GitHits, it used 45% fewer tokens and passed the tests on the 3rd attempt instead of the 8th: https://www.youtube.com/watch?v=0YmwLhH2Ohs
Install is:
npx githits@latest init
It works with coding agents either through the CLI directly or through the local MCP server that the CLI configures automatically.
I’m Olli-Pekka, co-founder of GitHits (and the creator of opencv-python package).
GitHits is a CLI and local MCP server that gives coding agents access to the open-source code their projects actually depend on.
Coding agents already have access to your local codebase. The missing context is the open-source code behind your dependencies.
When an agent hits a framework edge case, an undocumented integration, or version-specific behavior, the answer often lives in dependency source code, tests, changelogs, issues, or discussions rather than in the application code itself. Without access to that information, agents spend more time exploring dead ends, retrying approaches, and consuming tokens before converging on a solution.
During the private beta, GitHits focused on generating examples. You could describe a problem in natural language, and it would search open-source repositories, issues, discussions, and package metadata for similar implementations, then generate an example grounded in real code.
The public beta expands GitHits beyond example generation. Agents can now move beyond finding examples and inspect the dependencies they use directly: searching source code, accessing documentation, inspecting dependency graphs, vulnerabilities and changelogs, reviewing package upgrades, and retrieving information for the specific package versions used by a project.
No repo cloning is needed. We have built a version-aware index of open-source code and documentation. Popular projects are pre-indexed across ecosystems. If a package or repository is missing, we index it on demand, typically in 10-20 seconds for an average-sized project.
The CLI and local MCP server currently expose the following tools:
- get_example for implementation examples from open source
- search, code_grep, code_files, code_read for code navigation
- docs_list, docs_read for documentation access
- pkg_info, pkg_deps, pkg_vulns, pkg_changelog, pkg_upgrade_review for package inspection
Package navigation currently supports npm, PyPI, crates.io, Maven, NuGet, Go modules, RubyGems, Packagist, Hex, Zig, vcpkg and Swift.
Code navigation currently supports C++, C#, Dart, Elixir, Erlang, Go, Java, JavaScript, Kotlin, Lua, PHP, Python, R, Ruby, Rust, Scala, Swift, TypeScript and Zig. We also have more limited support for Bash, CSS, HTML, Markdown, Proto, SCSS, Svelte, Vue. We are constantly adding support for more ecosystems, and advanced navigation tools over our indexed code graph are also on the roadmap.
We benchmarked Codex on a real integration task with and without GitHits. With GitHits, it used 45% fewer tokens and passed the tests on the 3rd attempt instead of the 8th: https://www.youtube.com/watch?v=0YmwLhH2Ohs
Install is:
It works with coding agents either through the CLI directly or through the local MCP server that the CLI configures automatically.
Launching as well on Product Hunt today: https://www.producthunt.com/products/githits
Would love feedback, especially from people using coding agents heavily in real projects.