The problem

You want to contribute to open source. So you:

1. Scroll through hundreds of GitHub repos 😩
2. Find a "good first issue" — already taken 😤
3. Spend an hour understanding a dead repo 💀
4. Discover someone already opened a PR 😭
5. Give up, go back to tutorials 📺

This loop is broken. OpenCollab fixes it in one sentence to your AI assistant.

The fix

"Find me a good first issue I can contribute to this weekend."

Claude calls OpenCollab → scans your GitHub profile → picks your strongest language → finds beginner-friendly issues in active repos with no existing PR → hands you the issue + full context to draft the fix.

One sentence. A real mergeable issue.

📦 Install in 60 seconds

1. Get a free GitHub token

github.com/settings/tokens → Generate new token (classic) → check public_repo → copy.

2. Add to your AI tool

{
  "mcpServers": {
    "opencollab": {
      "command": "uvx",
      "args": ["opencollab-mcp"],
      "env": {
        "GITHUB_TOKEN": "your_github_token_here"
      }
    }
  }
}

{
  "mcpServers": {
    "opencollab": {
      "command": "uvx",
      "args": ["opencollab-mcp"],
      "env": {
        "GITHUB_TOKEN": "your_github_token_here"
      }
    }
  }
}

pip install opencollab-mcp

{
  "mcpServers": {
    "opencollab": {
      "command": "opencollab-mcp",
      "env": {
        "GITHUB_TOKEN": "your_github_token_here"
      }
    }
  }
}

docker build -t opencollab-mcp .
docker run -e GITHUB_TOKEN=ghp_xxx -p 8000:8000 opencollab-mcp

🎬 See it in action

The killer demo — 3 prompts to go from zero to a drafted PR

1️⃣ Analyze me

"My GitHub username is prakhar1605. Am I ready to contribute to open source?"

Readiness: 72/100. You know 4 languages, 15 public repos, haven't opened PRs yet. Tips: start with a docs fix, try a repo you already use.

2️⃣ Find me a mergeable issue

"Find me a Python good-first-issue I can finish in 1–2 hours. Make sure nobody's working on it."

Returns 5 issues · filters out ones with assignees or linked PRs · sorts by "quickness score" (short body, few comments, easy label).

3️⃣ Plan the PR

"Plan a PR for issue #456 in owner/repo."

Pulls the issue body, comments, CONTRIBUTING.md, the repo's directory structure, and the default branch — hands Claude everything needed to draft the actual code.

That's the whole loop: Analyze → Find → Plan → Ship.

More things you can just say

🛠️ All 22 tools

⚡ Why it's different

You ask Claude → Claude calls OpenCollab tools → Tools hit GitHub's free API → Data flows back → Claude reasons over it → You get a real, specific answer

OpenCollab is a data bridge, not an AI. Your AI assistant does the thinking. That means:

• 🆓 Zero AI costs — pure GitHub API, no paid services
• 🔑 No secrets besides a free GitHub token
• 💻 Runs locally on your machine (STDIO transport) or as a container (streamable-HTTP)
• 🔒 Private — your GitHub data never leaves your computer
• ⚡ Fast — direct API calls + 5-minute in-memory cache to soften rate-limit pressure
• 🧪 Tested — pytest suite + CI on every push

🏗️ Develop / Contribute

This project is itself a great first contribution target.

git clone https://github.com/prakhar1605/Opencollab-mcp.git
cd Opencollab-mcp
pip install -e ".[dev]"
export GITHUB_TOKEN="your_token_here"

# Run the server
python -m opencollab_mcp

# Run the tests
pytest

# Lint
ruff check src tests

# Or test interactively with the MCP Inspector:
npx @modelcontextprotocol/inspector python -m opencollab_mcp

Project layout

src/opencollab_mcp/
├── server.py          # entry point, transport selection
├── github_client.py   # cached HTTP wrapper with friendly errors
├── helpers.py         # days_ago, truncate, parse_issue_number, …
├── models.py          # Pydantic input models
├── constants.py       # all magic numbers / thresholds
└── tools/
    ├── discovery.py   # 6 matching tools
    ├── evaluation.py  # 7 scoring tools
    ├── issues.py      # 6 issue-intelligence tools
    └── profile.py     # 3 readiness tools

Check open issues labelled good first issue.

🗺️ Roadmap

• 22 tools shipped
• Published on PyPI — uvx opencollab-mcp works out of the box
• In-memory caching layer (fewer API calls, less rate-limit friction)
• GitHub Actions CI
• Pytest suite with httpx MockTransport
• Streamable-HTTP transport for remote deployment
• first_pr_generator — one-shot "find + plan + draft my first PR"
• track_my_prs — dashboard of your open PRs with staleness nudges
• skill_gap — compare your skills vs a target repo's stack

Got an idea? Open an issue — that's the fastest path in.

📄 License

MIT — see LICENSE.