Hey everyone! I'm working on a concept for an AI tool that could automate the process of solving GitHub issues. The workflow I'm imagining is as follows: I create an issue in a GitHub repo, and then the AI agent picks it up, generates the necessary code, and submits a pull request (PR). An AI reviewer would then evaluate the PR and provide feedback, after which the developer can make updates based on the review. Finally, once I approve the PR, the issue gets closed. I'm currently exploring tools like LangChain, OpenHands, and MCP, but I'm feeling a bit overwhelmed with how to start putting this all together. If anyone has experience with similar projects or can suggest frameworks that help with AI agent teamwork, I'd really love to hear your thoughts!
4 Answers
GitHub just launched their MCP, which seems like a solid place to start. It's specifically designed for managing PRs and issues, so it could fit well into your workflow.
It might be worth considering why you want a separate AI Reviewer. Couldn’t you have the review process happen before opening the PR? It could save some time and streamline things. Plus, seeing the agents' conversations could help with clarity.
Check out [janito.dev](http://janito.dev) for some coding insights. It would be great to see community contributions for GitHub tools there!
I’m actually working on something quite similar! I’m using Ollama for the LLM and connecting to it using Go. Although LangChain for Go is decent, the Python version feels more complicated. It’s pretty easy to integrate any LLM you choose, just keep things straightforward and avoid unnecessary layers. I suggest using Git client libraries for GitHub interactions.
What kind of branching strategy are you considering? Do you have a specific flow in mind yet?
That's a good point! It makes sense to have the review happen constantly instead of waiting until after the PR is made.