Reviewing your agents' work locally with G

The other day, while talking with Marek about how we use agentic coding tools in our daily workflow, I realized that the pattern of opening a draft pull request just to use GitHub's PR review interface might be more common among developers than I initially thought. This practice of leveraging GitHub's review UI to systematically review work, leave comments, and mark files as viewed or not viewed has become second nature to many of us. I particularly appreciate GitHub's user interface and the way it incorporates features that transform the review experience into something resembling a checklist. This checklist-like approach is especially handy because it allows me to clearly flag whether a piece of work has already been reviewed by me, preventing duplicate effort and ensuring nothing slips through the cracks.

This conversation sparked an idea: what if we could bring this same powerful reviewing experience into a local development environment where it works completely offline? Better yet, what if it could be integrated as an MCP server that plugs directly into agentic coding tools like Claude Code or Codex, enabling these tools to read and resolve comments left during the review process? That's exactly how Guck came to life. The name Guck is derived from the German verb "gucken," which means to take a look or to have a glance at something. This etymology perfectly captures what Guck is all about: it's a tool designed for both you and AI agents to collaboratively review the work that has been done and iterate on it together in a seamless manner.

Guck is a command-line interface tool that you install on your system and configure in your shell environment. Once properly configured, it intelligently starts and stops a server automatically as you navigate into or out of directories that contain a Git repository. This automatic behavior means you don't have to remember to manually start or stop anything. When you're working inside one of these directories, you can simply run the guck command, and it will instantly open a web interface in your browser where you can begin reviewing your work. The graphical user interface supports marking individual files as viewed (similar to GitHub's functionality), and it also allows you to leave detailed comments on specific lines of code. These comments can then be read and acted upon by AI agents using the MCP protocol, creating a true collaborative workflow between human and machine. The state of your reviews, including which files you've marked as viewed and all the comments you've left, is persisted across Guck sessions on your system. This persistence is intelligently associated with the specific repository, branch, commit, and even the exact line number where comments were made, ensuring that your review context is never lost. Pretty cool, isn't it?

The project is open-source and released under the MIT license, which means you're free to use it, modify it, and adapt it to fit your specific needs. I'd love for you to give it a shot and see how it fits into your development workflow. If you find it useful or have ideas on how to make it even better, I'd genuinely appreciate your contributions. Whether you want to open a pull request with improvements, report bugs, suggest new features, or simply share your thoughts through GitHub issues, all feedback is welcome. The beauty of open-source is that we can build and improve tools like this together as a community, making them more robust and useful for everyone who relies on them.