PRismatic: A Code Review Agent

An AI Agent for reviewing GitHub pull requests from multiple perspectives, built using the mcp-agent framework.

Features

PRismatic analyzes pull requests from four specialized perspectives:

1. Security Reviewer: Identifies security vulnerabilities and best practices
2. Performance Reviewer: Focuses on performance optimizations and bottlenecks
3. Clarity Reviewer: Evaluates code readability, naming conventions, and maintainability
4. QA Reviewer: Assesses test coverage and testing best practices

Each reviewer provides specialized feedback, which is then aggregated into a comprehensive review report with prioritized recommendations.

MCP Integration

Each reviewer utilizes the GitHub and FileSystem MCP Servers to access PR metadata, diff information, and other reviewers' comments. This integration allows for a comprehensive analysis by exploring the local repository to provide deeper context for the changes.

Setup

1. Install uv if you haven't already

2. Create a mcp_agent.secrets.yaml file (using the .example file as a template) with your API keys.

Usage

To review a GitHub pull request:

uv run main.py https://github.com/username/repo/pull/123

Configuration

GitHub Access Token

You need a GitHub Personal Access Token

• Go to Personal access tokens (in GitHub Settings > Developer settings)
• Select which repositories you'd like this token to have access to (Public, All, or Select)
• Copy the generated token

Anthropic API Key

You need an Anthropic API Key

• Go to Anthropic API Console
• Copy your API key

Requirements

• Python 3.11+
• Anthropic API key
• GitHub access for PR details