GitHub PR Review Context MCP

AI code review that remembers how YOUR team reviews code.

Most AI code reviewers have zero memory. They don't know your team flagged async/await misuse in 12 PRs last quarter. They don't know you've rejected the same error-handling pattern 5 times. This MCP fixes that.

It indexes your repo's PR history and gives your AI coding tool the institutional memory to review code the way your team actually does.

Demo

demo

"Review this diff using our team's review history" → finds past review comments, flags patterns your team has rejected before.

Who Is This For

Open Source Contributors

You want to contribute to psf/black or vercel/next.js but don't know what the maintainers care about. The docs don't tell you that the team hates a certain pattern or that they've rejected the same approach 5 times. This tool reads 60 PRs of real review history and tells you exactly that — before you click "Create Pull Request."

Junior Developers Joining a Team

You're new. You don't know why the team always asks for a certain error handling style or why they keep rejecting a particular abstraction. This tool reads your team's private PR history on day one and gives you the institutional memory that usually takes 6 months to build.

Engineering Leads Tired of Repeating Themselves

You've left the same review comment 15 times. This tool learns from your past comments and catches the same issues automatically, before PRs hit human review.

Developer Tool Builders

You want context-aware code review in your CI pipeline, IDE plugin, or internal dev platform. This MCP is a composable component you can plug in — any LLM, any client that supports MCP.

Quickstart

Prerequisites

Python 3.10+
GitHub Personal Access Token — repo scope for private repos, public_repo for public only
An LLM API key — multiple providers supported, see Configuring Your LLM

Install

git clone https://github.com/paarths-collab/github-pr-context-mcp
cd github-pr-context-mcp
pip install -r requirements.txt
cp .env.example .env
# Fill in GITHUB_TOKEN + your chosen LLM key

How to Use It

Zero-setup (recommended)

Just talk to your AI tool. No manual indexing needed:

"I want to contribute to psf/black. Can you review this code using their history?"

The server will:

Check if psf/black is already indexed locally
If not — ask whether you want to store it permanently (disk, reusable) or temporarily (this session only, no disk usage)
Fetch and index the repo automatically
Review your code using real past PR patterns

Storage: Permanent vs Temporary

| | Permanent 💾 | Temporary ⚡ | |---|---|---| | Stored | Disk (ChromaDB) | RAM only | | Survives restart | ✅ Yes | ❌ No | | Disk usage | ~5–20 MB per repo | 0 MB | | Best for | Repos you query often | One-off exploration |

Manual indexing (optional)

Pre-index before connecting to your AI tool:

python indexer.py psf/black --pages 2
# Or use full URLs: python indexer.py https://github.com/psf/black
# --pages: 1-10, default 2. More pages = better context but slower.

Example prompts

"Review this code using the psf/black repo's history"
"What does the vercel/next.js team commonly flag in reviews?"
"Switch to https://github.com/tiangolo/fastapi context"
"Find past review comments about error handling in my active repo"
"List all repos I've indexed"

Configuring Your LLM

Set LLM_PROVIDER and LLM_MODEL in your .env. No lock-in.

| Provider | LLM_PROVIDER | Example LLM_MODEL | Cost / Setup | |---|---|---|---| | Cerebras | cerebras | llama3.1-8b | Free tier — cloud.cerebras.ai | | Groq | groq | llama-3.3-70b-versatile | Free tier — console.groq.com | | Gemini | gemini | gemini-2.5-flash | Free tier — aistudio.google.com | | OpenAI | openai | gpt-5 | Paid (Recommended for highest quality) | | Anthropic | anthropic | claude-sonnet-4-6 | Paid (Recommended for highest quality) | | Ollama | ollama | qwen2.5:7b | 100% Free & Local |

(Tip: While free tiers are great for testing the waters, code review benefits massively from complex reasoning. If available, use a paid frontier model like Claude Sonnet 4.6 or GPT-5 for production use.)

Set your key as LLM_API_KEY in .env — works for any provider:

Integrations

All integrations use the same MCP stdio transport. Replace /absolute/path/to with your actual path.

Antigravity (Google)

Mac/Linux: ~/.gemini/antigravity/mcp_config.json
Windows: %APPDATA%\.gemini\antigravity\mcp_config.json

{
  "mcpServers": {
    "github-pr-context": {
      "command": "python",
      "args": ["/absolute/path/to/github-pr-context-mcp/server.py"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token",
        "LLM_PROVIDER": "cerebras",
        "LLM_MODEL": "llama3.1-8b",
        "LLM_API_KEY": "your_key",
        "CHROMA_PERSIST_DIR": "/absolute/path/to/chroma_db"
      }
    }
  }
}

Claude Desktop

Mac: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "github-pr-context": {
      "command": "python",
      "args": ["/absolute/path/to/github-pr-context-mcp/server.py"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token",
        "LLM_PROVIDER": "cerebras",
        "LLM_MODEL": "llama3.1-8b",
        "LLM_API_KEY": "your_key",
        "CHROMA_PERSIST_DIR": "/absolute/path/to/chroma_db"
      }
    }
  }
}

Claude Code

In your project root or home directory, create .mcp.json:

{
  "mcpServers": {
    "github-pr-context": {
      "command": "python",
      "args": ["/absolute/path/to/github-pr-context-mcp/server.py"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token",
        "LLM_PROVIDER": "cerebras",
        "LLM_MODEL": "llama3.1-8b",
        "LLM_API_KEY": "your_key",
        "CHROMA_PERSIST_DIR": "/absolute/path/to/chroma_db"
      }
    }
  }
}

Cursor

Mac/Linux: ~/.cursor/mcp.json
Windows: %APPDATA%\Cursor\mcp.json

Or project-level: .cursor/mcp.json in your repo root.

{
  "mcpServers": {
    "github-pr-context": {
      "command": "python",
      "args": ["/absolute/path/to/github-pr-context-mcp/server.py"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token",
        "LLM_PROVIDER": "cerebras",
        "LLM_MODEL": "llama3.1-8b",
        "LLM_API_KEY": "your_key",
        "CHROMA_PERSIST_DIR": "/absolute/path/to/chroma_db"
      }
    }
  }
}

Then: Cursor Settings → Features → MCP → enable.

Windsurf

Mac/Linux: ~/.codeium/windsurf/mcp_config.json
Windows: %APPDATA%\Codeium\windsurf\mcp_config.json

{
  "mcpServers": {
    "github-pr-context": {
      "command": "python",
      "args": ["/absolute/path/to/github-pr-context-mcp/server.py"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token",
        "LLM_PROVIDER": "cerebras",
        "LLM_MODEL": "llama3.1-8b",
        "LLM_API_KEY": "your_key",
        "CHROMA_PERSIST_DIR": "/absolute/path/to/chroma_db"
      }
    }
  }
}

GitHub Copilot (VS Code)

Create .vscode/mcp.json in your workspace:

{
  "servers": {
    "github-pr-context": {
      "type": "stdio",
      "command": "python",
      "args": ["/absolute/path/to/github-pr-context-mcp/server.py"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token",
        "LLM_PROVIDER": "cerebras",
        "LLM_MODEL": "llama3.1-8b",
        "LLM_API_KEY": "your_key",
        "CHROMA_PERSIST_DIR": "/absolute/path/to/chroma_db"
      }
    }
  }
}

Enable: VS Code Settings → chat.mcp.enabled: true. Requires VS Code 1.99+ with Copilot Chat.

OpenCode

Mac/Linux: ~/.config/opencode/config.json
Windows: %APPDATA%\opencode\config.json

{
  "mcp": {
    "github-pr-context": {
      "command": "python",
      "args": ["/absolute/path/to/github-pr-context-mcp/server.py"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token",
        "LLM_PROVIDER": "cerebras",
        "LLM_MODEL": "llama3.1-8b",
        "LLM_API_KEY": "your_key",
        "CHROMA_PERSIST_DIR": "/absolute/path/to/chroma_db"
      }
    }
  }
}

MCP Tools

| Tool | What It Does | |---|---| | ensure_repo_ready | Smart loader — checks local DB, prompts for storage choice if new, fetches + indexes automatically | | set_active_repo | Switch context to a different already-indexed repo | | list_indexed_repos | Show all repos indexed locally with storage type and doc count | | semantic_search_reviews | Find past review comments similar to a code snippet or concept | | review_code_with_history | Full AI review grounded in this repo's real PR history | | get_team_review_patterns | Summarize top patterns this team flags in reviews | | get_index_stats | Check document count and storage type for a repo |

Project Structure

github-pr-context-mcp/
├── server.py              # MCP server — tool definitions + session state
├── indexer.py             # CLI: pre-index a repo without an AI tool
│
├── fetcher/               # GitHub API layer
│   ├── client.py
│   ├── queries.py
│   └── transform.py
│
├── storage/               # Vector DB layer
│   ├── encoder.py
│   ├── document_builder.py
│   └── vector_store.py
│
└── inference/             # LLM inference layer
    ├── providers.py       # Cerebras / Groq / Gemini / OpenAI / Anthropic / Ollama
    └── review.py

Roadmap

[x] GraphQL fetching with review thread flattening
[x] Local embeddings with ChromaDB persistence (permanent + temporary storage)
[x] Swappable LLM backend (Cerebras / Groq / Gemini / OpenAI / Anthropic / Ollama)
[x] Smart repo loader — no manual steps needed from the AI tool
[x] Session memory — per-repo context switching
[x] Integrations: Antigravity, Claude Desktop, Claude Code, Cursor, Windsurf, GitHub Copilot, OpenCode
[ ] Incremental indexing
[ ] File-aware retrieval
[ ] Webhook auto-indexing
[ ] Streaming inference

License

MIT — use it, fork it, build on it.

If this saved you a PR rejection, a ⭐ helps others find it.

MCP Servers

GitHub PR Review Context MCP

Demo

Who Is This For

Open Source Contributors

Junior Developers Joining a Team

Engineering Leads Tired of Repeating Themselves

Developer Tool Builders

Quickstart

Prerequisites

Install

How to Use It

Zero-setup (recommended)

Storage: Permanent vs Temporary

Manual indexing (optional)

Example prompts

Configuring Your LLM

Integrations

Antigravity (Google)

Claude Desktop

Claude Code

Cursor

Windsurf

GitHub Copilot (VS Code)

OpenCode

MCP Tools

Project Structure

Roadmap

License

安装包（如果需要）

Cursor 配置 (mcp.json)

GitHub PR Review Context MCP

Demo

Who Is This For

Open Source Contributors

Junior Developers Joining a Team

Engineering Leads Tired of Repeating Themselves

Developer Tool Builders

Quickstart

Prerequisites

Install

How to Use It

Zero-setup (recommended)

Storage: Permanent vs Temporary

Manual indexing (optional)

Example prompts

Configuring Your LLM

Integrations

Antigravity (Google)

Claude Desktop

Claude Code

Cursor

Windsurf

GitHub Copilot (VS Code)

OpenCode

MCP Tools

Project Structure

Roadmap

License

安装包 （如果需要）

Cursor 配置 (mcp.json)

安装包（如果需要）