You know the feeling.
Someone drops a GitHub link in Discord. Or you find a repo on HN. Or your team says "hey, we're migrating to this library."
You open it. Stars look good. But then:
- README is either missing or 3 years old
- 47 top-level files staring at you
- No idea where the actual logic lives
- You clone it. Browse it. Still confused 10 minutes later.
This happened to me constantly. And every time I tried to give a repo to Claude Code or Cursor as context, I'd end up dumping random files and hoping for the best.
So I built something to fix it.
What I built: RepoContext
RepoContext — paste any GitHub URL, get a structured brief in seconds.
No AI. No signups. No agents. Fully deterministic.
Here's what it actually gives you:
1. Onboarding Brief
Not just a README summary. It surfaces:
- What type of project this is
- The files most worth reading first (ranked by importance)
- Dependency risk signals
- Repo health: open issues, PR pressure, activity status
2. Full File Tree
Browse the complete tree — up to 28k+ entries. Click any file to get an inline preview (first 16KB, no download needed).
3. Export for LLM
One click. You get a structured Markdown export ready to paste into Claude Code, Cursor, ChatGPT, Gemini CLI, or OpenClaw.
No more guessing which files to include. No more dumping the entire repo and hoping the context window survives.
4. Bulk File Download
Select the files you actually need. Download them as a ZIP. Useful when you want to give a specific subsystem to an AI tool without cloning the full repo.
5. Private Repo Support
Add your GitHub PAT in the input. Works with private repos and gives you higher rate limits on the GitHub API.
The workflow before vs after
Before:
- Open repo
- Stare at 40+ files
- Start randomly clicking around
- Clone it just to understand it
- Paste random files into Claude
- Get mediocre output because context was incomplete
After:
- Paste URL into RepoContext
- Read the brief (30 seconds)
- Know exactly which files matter
- Export structured context for your AI tool of choice
- Get better output because the context is actually structured
Why no AI?
Intentional.
AI-generated summaries can hallucinate. When you're evaluating a codebase for production use or preparing context for another AI tool, you want deterministic output — the actual structure, the actual metadata, the actual health signals.
RepoContext uses heuristics, not inference. What you see is what's actually in the repo.
Try it
→ repocontext.ajaymathuriya.com
Paste any public GitHub URL. Takes about 3 seconds.
If you try it on a repo you know well, you'll immediately see whether the onboarding brief matches your mental model. That's always the most satisfying part.
It's open source
→ github.com/ajaycodesitbetter/RepoContext
Built with Next.js 14, TypeScript, Tailwind v4, deployed on Vercel.
PRs welcome. If you find a repo where the file ranking is wrong or the health signals look off, open an issue — I want to know.
Built this as a student developer. If it saves you 20 minutes on your next codebase evaluation, that's the whole point.