Code Review MCP Server

Status: archived / portfolio reference. Built March 2026 as a focused exploration of FastMCP + AST-based code-quality heuristics. The deterministic tools (Ruff / ShellCheck / ESLint wrappers, secret-pattern checks) still work; the LLM-style review feedback is largely overlapped by modern coding assistants, which is why this is frozen rather than actively maintained. Fork if you want to extend.

An open-source Model Context Protocol (MCP) server that brings senior-level code review into your editor. Use it with Cursor or any MCP client to get quality checks, refactor suggestions, security checks, and best-practice guidance as you code.

Quick start

git clone <this-repo>
cd mcp_server
python -m venv venv
./venv/bin/pip install -r requirements.txt
./venv/bin/python code_review_mcp_server.py

With Cursor: Add the server to your MCP config (e.g. copy mcp.json into ~/.cursor/ and set workingDirectory to this repo). Cursor will then offer tools like senior_review, review_code_quality, and security_review when you work on code.

What it does

The server exposes tools over MCP that your editor can call to:

Area	Tools
One-shot review	senior_review — checklist and concrete suggestions (naming, errors, types, tests, security, DRY)
Quality	review_code_quality — long functions, nesting, type hints, error handling
Security	security_review — eval/exec, shell/SQL injection, hardcoded secrets, permissions
Refactor	refactor_code — split functions, reduce complexity, unused imports, naming
Structure	suggest_code_split, suggest_folder_structure, suggest_reuse — split by logic, folder layout, reuse existing code
Tests	generate_tests — scenarios and edge cases per function
Static analysis	Ruff (Python), ShellCheck (Bash), ESLint (JS/TS), patch generation

So instead of "quick AI code," you get feedback that matches what a senior engineer would expect in a code review: clear structure, fewer security risks, and maintainable patterns.

Requirements

• Python 3.10+
• Optional: Ruff for Python linting (pip install ruff), ShellCheck for Bash, ESLint (e.g. via npx) for JavaScript/TypeScript

Project structure

mcp_server/
  main.py                      # Minimal entry point
  code_review_mcp_server.py    # Entry point with config and logging
  tools/                        # MCP tools (quality, security, refactor, etc.)
  utils/                        # Helpers (temp files, diffs)
  tests/                        # Unit tests (tools, utils, common)
  mcp.json                      # Example MCP config for Cursor
  requirements.txt
  requirements-dev.txt          # Dev deps (pytest); optional
  pyproject.toml               # Project metadata and pytest config

Testing

From the project root (with the venv activated and deps installed):

python -m unittest discover -s tests -p 'test_*.py' -v

Or install dev deps and use pytest: pip install -r requirements-dev.txt then pytest tests/ -v.

Usage

1. Run the server from the project directory:

   ./venv/bin/python code_review_mcp_server.py

   or python main.py (both use the same config and logging).

2. Use from Cursor: Point your Cursor MCP config at this repo. The example mcp.json uses relative paths: workingDirectory should resolve to the cloned repo (e.g. ../mcp_server if the config file lives in ~/.cursor). For reliability, you can set workingDirectory to the absolute path of this repo (e.g. ~/mcp_server).

Getting better results

• Pass file_path when calling senior_review, review_code_quality, or security_review. Findings will include file:line references so you can jump to the exact location.
• Use focus with senior_review to narrow the checklist: "security" (secrets, injection, permissions), "api" (naming, types, docs), or omit for the full checklist.
• Review in small chunks. Run review on one file or one concern at a time; large blobs of code produce noisier or vaguer feedback.
• Ask for one thing at a time. For example: “Run security_review on this file” or “Run senior_review with focus=api on this function.”

License and author

License: MIT — see LICENSE.
Author: Dmitry Troshenkov.

Contributions and feedback are welcome.