🚀 GitHub PR Context Tools

This repository provides two ways to leverage your GitHub PR history in your AI workflows. Choose the tool version that matches your architecture.

💎 GitHub PR Context (v0.3.0)

The Hardened Pure Context Retrieval Engine.

Designed for users who want maximum speed and reliability by offloading reasoning to their IDE agent (Cursor, Claude Desktop, Windsurf). It focuses exclusively on high-performance retrieval and data pruning.

✨ Key Features (v0.3.0):

• Asynchronous Hardening: Built with tenacity retries and non-blocking I/O for heavy repository indexing.
• Optimized for IDE Agents: Prunes redundant metadata and strips useless fields to save token space in your prompt.
• Modern Infrastructure: Uses SQLite-backed cursors for progress tracking and global thread locks for data safety.
• Pure Context: No built-in LLM dependencies; feeds raw, high-quality historical patterns directly to your primary agent.

🛠 Installation

pipx install github-pr-context-mcp

🏛 Legacy Context Agent (v0.2.9)

The Original "All-in-One" Inference Agent.

Use this version if you still rely on the server's internal LLM provider logic (OpenAI/Anthropic keys configured in the server).

🛠 Installation

pipx install github-pr-context-mcp==0.2.9

Context layer for AI code review, grounded in your repository's real pull request history.

❤️ Community Feedback

We want to hear from you! Your reviews help us improve and reach more developers.

👉 Write a Review / Share your Experience

"This MCP is like having a teammate with perfect memory of every PR ever written." — Example Review

Overview

GitHub PR Review Context MCP gives AI assistants institutional review memory.

Instead of generic feedback, reviews are informed by historical reviewer comments, recurring quality patterns, and repository-specific standards from your own PR history.

Core Value

• Improves review consistency across teams and repositories.
• Reduces repeated reviewer feedback on known issues.
• Integrates with any MCP-compatible client and multiple LLM providers.

Key Capabilities

🚀 Quick Start

🚀 Zero-Setup (uvx / pipx / npx)

The fastest way to use the server. No cloning required. Just run one of these commands directly in your terminal or use them in your IDE's MCP settings:

[!TIP] Don't clone this repo to get AI rules! Once installed, run generate_repo_rules inside YOUR project to automatically create .cursorrules or CLAUDE.md tailored to your own team's PR history.

Using uvx (Recommended for speed):

uvx github-pr-context-mcp

Using pipx (Recommended for stability):

pipx run github-pr-context-mcp
# Or install permanently:
pipx install github-pr-context-mcp

Using npx (Smithery bridge):

npx -y @smithery/cli run github-pr-context-mcp

⚠️ Manual Installation (Git Clone / Advanced)

[!WARNING] Running from a git clone is only recommended for developers contributing to this project. For general use, please use the pipx method above.

If you have cloned the repository for development:

1. Create a virtual environment: python -m venv .venv
2. Activate it and install: pip install -e .
3. Run automatic setup: python scripts/install_clients.py

For full configuration (Cursor, Claude Desktop), see the Quick Start Guide.

[!IMPORTANT] 🚀 USE THE OFFICIAL PACKAGE: This project is now on PyPI. To ensure seamless updates and zero configuration friction, do NOT git clone.

Recommended Install:

pipx install github-pr-context-mcp

Or run instantly with: uvx github-pr-context-mcp

🛠️ Usage Modes: Solo vs. Team

This MCP server is built to scale from a single machine to an entire engineering organization.

👤 Solo Developer (Local Mode)

Best for: Privacy, local-first control, and zero hosting costs.

• How it works: Run via uvx, pipx, or a local git clone.
• Storage: ChromaDB stays on your local machine.
• Security: Your GitHub Token and LLM keys never leave your device.
• Setup: See Quick Start.

🤝 Team Collaboration (Hosted Mode — UPCOMING)

Best for: Scaling team-wide PR standards and centralized infra.

• How it works: One deployment on Render (Coming Soon) shared by the whole team.
• Isolation: Strict Gmail-based namespace isolation (driven by SQLite). User A's indexed data is mathematically invisible to User B.
• Economics: Pooled LLM credits and a single shared indexing server.
• Setup: See Deployment Guide.

🌟 Zero-Friction Setup (Upcoming)

If your team has hosted this MCP on Render, you do NOT need to git clone or install anything. Just drop this snippet into your IDE:

"github-pr-context": {
  "type": "sse",
  "url": "https://YOUR-RENDER-URL.onrender.com/mcp",
  "headers": {
    "Authorization": "Bearer YOUR_TOKEN"
  }
}

That's it. If your IDE supports native MCP SSE connections, you are immediately connected to the secure Render deployment.

🧰 Core Tools Reference

The server exposes 12 core tools for IDE agents and developers. For a deep dive on when to use each, see the Tool Strategy Guide.

📖 Documentation

Detailed guides for deep dives and specific configurations:

• 🛠️ Quick Start & Usage — Setup and basic commands.
• ⚙️ LLM Configuration — Switching between OpenAI, Anthropic, Gemini, and Cerebras.
• 🧩 Tool Strategy & Selection Guide — When to use which tool (for humans and agents).
• 🏗️ Architecture & Pipeline — How the RAG engine and indexing work.
• 🔌 Integrations — Connecting to Cursor, Claude Desktop, and more.

🛠️ Troubleshooting

• "command not found": Use absolute paths in your configuration. Run github-pr-context-mcp config to get your exact path.
• "PermissionError: [WinError 32]": The binary is locked by a running process. Close Claude/Cursor, run taskkill /F /IM github-pr-context-mcp.exe, then retry the upgrade.
• Rate Limit Errors: Ensure your GITHUB_TOKEN is valid and has repo scope.

📣 Community & Feedback

We want to hear from you — whether you are a solo developer or a team at a large company!

• Feedback: Please open an issue or start a discussion if you have ideas or encounter bugs.
• Star ⭐: If this tool saves you time, give it a star! It helps others find the project.
• Corporate: Is your team using this? Join our "Adopters" list by opening a PR to add your team's name.

⚖️ License

MIT