Code Reviewer MCP Server

An MCP (Model Context Protocol) server that performs automated code reviews based on customizable reviewer personas. Integrates seamlessly with Cursor IDE and other MCP-compatible tools.

Overview

This server provides automated code review capabilities by analyzing git diffs against configurable review standards. It supports custom reviewer personas, making it easy to enforce team-specific code quality standards.

Features

The server exposes 7 tools:

Tool	Description
get_branch_diff	Get git diff between current branch and base branch
get_changed_files	List files changed on current branch with stats
review_diff	Get diff with review context and persona standards
review_file	Review a specific file against standards
get_persona	View the active reviewer persona
get_review_checklist	Get the full review checklist
generate_review_report	Generate a markdown review report file

Custom Persona Support

All review tools accept a persona_file parameter to use a custom reviewer persona:

persona_file: "personas/example_persona.md"

Default locations (checked in order):

1. personas/example_persona.md in the MCP server directory
2. notebooks/code_reviewer_persona.md in your project root (for backward compatibility)
3. Embedded default persona (if no file found)

Example persona: See personas/example_persona.md for a complete example of a reviewer persona.

Installation

Prerequisites

• Python 3.10+
• uv (recommended) or pip

Setup

1. Install dependencies:

   cd code-reviewer-mcp
   uv sync  # or: pip install -r requirements.txt

2. Set up the Cursor rule (optional but recommended):

   Copy the template rule to your project's .cursor/rules/ directory:

   mkdir -p .cursor/rules
   cp tools/code-reviewer-mcp/.cursor/rules/code-review.mdc.template .cursor/rules/code-review.mdc

   Then customize it for your project's review standards.

3. Configure in Cursor by adding to ~/.cursor/mcp.json:

   {
     "mcpServers": {
       "code-reviewer": {
         "command": "uv",
         "args": [
           "--directory",
           "/path/to/code-reviewer-mcp",
           "run",
           "server.py"
         ]
       }
     }
   }

   Or if using pip:

   {
     "mcpServers": {
       "code-reviewer": {
         "command": "python",
         "args": [
           "/path/to/code-reviewer-mcp/server.py"
         ]
       }
     }
   }

Architecture

The code reviewer consists of two components:

1. MCP Server (server.py): Provides the tools (get_branch_diff, review_diff, etc.)
2. Cursor Rule (.cursor/rules/code-review.mdc): Provides workflow instructions and review standards

The MCP server is reusable across projects - it provides generic code review tools.
The Cursor rule is project-specific - it defines your team's review standards and workflow.

When you ask for a code review, the Cursor rule instructs the AI on:

• Which MCP tools to call and in what order
• What standards to check against
• How to format the output

Usage in Cursor

Quick Start

1. Restart Cursor after installation to load the new MCP server
2. Ask Claude to review your code:
   • "Review my current branch"
   • "Review this PR against development"
   • "Check this file for issues"

Example Commands

Basic usage (uses default persona):

"Review the changes on my branch"
"Get the diff against development"
"Review src/my_module/file.py"
"Generate a code review report"

With custom persona (using @ reference):

"Review my code using @personas/example_persona.md"
"Review this file using the persona at @path/to/strict_reviewer.md"
"Generate a review report with @personas/example_persona.md"

The @file syntax in Cursor expands the file reference, making it easy to select different reviewer personas for different review styles.

Using the Cursor Rule

A Cursor rule at .cursor/rules/code-review.mdc automatically triggers the reviewer when you say "review", "code review", or "PR review".

What the rule does:

• Provides step-by-step workflow instructions for using the MCP tools
• Defines the review standards and checklist (type safety, documentation, etc.)
• Specifies the output format for reviews
• Handles persona file selection via @ syntax

For this project: The rule is located at .cursor/rules/code-review.mdc in the repo root.

For other projects: A template rule file is included at tools/code-reviewer-mcp/.cursor/rules/code-review.mdc.template. Copy it to your project's .cursor/rules/ directory and customize it for your team's standards.

The rule provides:

• Workflow instructions: Step-by-step guide on how to use the MCP tools
• Review standards: Checklist of what to check (can be customized per project)
• Output format: Structure for review comments
• Integration guidance: How to combine with Bitbucket MCP for PR comments

Persona Files

How Persona Selection Works

1. Explicit selection: Pass persona_file parameter with the path
2. Default locations (checked in order):
   • personas/example_persona.md in the MCP server directory
   • notebooks/code_reviewer_persona.md in your project root (for backward compatibility)
3. Embedded fallback: Uses built-in persona if no file found

Example persona: See personas/example_persona.md for a complete example based on real code review patterns.

Creating a Custom Persona

Create a markdown file with your review standards. Example structure:

# Code Reviewer Persona: [Name]

## Review Philosophy
[Your approach to code review]

## Key Standards
### Type Safety
- [Your type checking rules]

### Documentation
- [Your documentation requirements]

### Code Style
- [Your style preferences]

## Common Callouts
- "Missing type hint" → Add type annotations
- "No tests" → Add test coverage

Switching Personas

You can have multiple persona files for different contexts. Store them in the personas/ directory:

• personas/example_persona.md - Example persona (included with this repo)
• personas/strict_reviewer.md - For production code (create your own)
• personas/junior_friendly.md - Educational, more explanatory (create your own)
• personas/security_focused.md - Emphasis on security patterns (create your own)

Start with personas/example_persona.md and customize it for your team's needs.

Default Review Standards

The embedded default persona checks for:

Type Safety

• Complete type hints on all functions
• Modern syntax (str | None over Optional)
• No Any types without justification

Documentation

• File headers with copyright
• Complete docstrings with Args/Returns

Code Organization

• Absolute imports only
• Magic numbers as constants
• Unused code removed

Error Handling

• Specific exceptions only
• Edge cases handled

Architecture

• Layer separation maintained
• Common logic in templates

Utilities

Export PR Comments

The utils/export_comments.py script helps export your Bitbucket PR comments to CSV for analysis or building training data for code review personas.

Usage:

# Set environment variables (recommended)
export ATLASSIAN_EMAIL="your-email@example.com"
export BITBUCKET_API_TOKEN="your-api-token"
export BITBUCKET_WORKSPACE="your-workspace"
export BITBUCKET_REPO_SLUG="your-repo"
export BITBUCKET_ACCOUNT_ID="your-account-id"  # Optional: filter to your comments only

# Export comments (from the code-reviewer-mcp directory)
python utils/export_comments.py

# Export only your comments (with account ID filter)
python utils/export_comments.py --account-id your-account-id

# Export all comments (no filter)
python utils/export_comments.py --account-id ""

# Or pass everything as arguments
python utils/export_comments.py \
  --email your-email@example.com \
  --token your-token \
  --workspace your-workspace \
  --repo your-repo \
  --output my_comments.csv \
  --account-id your-account-id

Note: This utility requires the requests library. Install with:

pip install requests
# or
uv add requests

Output: The script generates a CSV file with columns:

• pr_id, pr_title, pr_url
• comment_id, content
• file_path, line (for inline comments)
• created_on, updated_on

Development

Testing the Server

cd code-reviewer-mcp
uv run server.py

The server communicates via stdio, so you'll see it waiting for JSON-RPC messages.

Modifying the Persona

The reviewer persona is embedded in server.py in the REVIEWER_PERSONA constant. Update this to change review standards.

Limitations

• No inline comments: Cursor doesn't have an API to programmatically add inline comments to files. The server outputs reviews with file:line references that you can navigate to.

• Python-focused: Currently filters for *.py files by default. The file_filter parameter can be changed to include other file types.

Troubleshooting

Server not appearing in Cursor

1. Check ~/.cursor/mcp.json has the correct path
2. Restart Cursor completely (Cmd+Q on macOS)
3. Check the MCP logs: ~/Library/Logs/Claude/mcp*.log

Git errors

Ensure you're in a git repository when using the diff-related tools. The server needs access to git commands.

Optional: Bitbucket Integration

For teams using Bitbucket, you can optionally configure the @lexmata/bitbucket-mcp server to enable programmatic PR comment creation. This allows you to post review comments directly to Bitbucket pull requests.

Setting up Bitbucket MCP

1. Install the Bitbucket MCP server (if not already installed):

   npm install -g @lexmata/bitbucket-mcp

2. Configure in ~/.cursor/mcp.json:

   {
     "mcpServers": {
       "code-reviewer": {
         "command": "uv",
         "args": ["--directory", "/path/to/code-reviewer-mcp", "run", "server.py"]
       },
       "bitbucket": {
         "command": "npx",
         "args": ["-y", "@lexmata/bitbucket-mcp"],
         "env": {
           "BITBUCKET_WORKSPACE": "your-workspace",
           "BITBUCKET_REPO_SLUG": "your-repo",
           "BITBUCKET_APP_PASSWORD": "your-app-password"
         }
       }
     }
   }

3. Usage: Once configured, you can use Bitbucket MCP tools alongside the code reviewer:

   • Create PR comments programmatically
   • Fetch PR details
   • Post review feedback directly to Bitbucket

   Example workflow:

   1. Use code-reviewer tools to generate review feedback
   2. Use bitbucket-mcp tools to post comments to the PR
   

   Note: This integration is optional. The code reviewer works perfectly fine without it, generating review reports that you can manually copy to PR comments.