mcp-code-review 0.3.0

MCP server for LLM-powered code review — GitHub PRs, GitLab MRs, local diffs

mcp-code-review

An MCP server that brings structured code review into any MCP-compatible client. Review GitHub Pull Requests, GitLab Merge Requests, and local git diffs — with static analysis, project conventions, and multi-language report output.

---

Features

• Review GitHub PRs and GitLab MRs by URL
• Review local git diffs against any base branch
• Review individual files with static analysis
• Scan entire projects for issues
• Post review comments back to GitHub PRs and GitLab MRs
• Static analysis via Ruff (linting) and Bandit (security)
• Project conventions loaded from .codereview.yml or auto-detected
• Multi-language report output: English, Italian, Spanish
• Configurable review focus: all, security, performance, quality
• Saves review reports as Markdown files
• Professional engineering checklist with 13 analysis categories (SOLID, architecture, security, performance, data layer, observability, concurrency, and more)

---

Installation

Install the base package:

pip install mcp-code-review

Install with Python static analysis tools (Ruff + Bandit):

pip install "mcp-code-review[python]"

---

Quick Start

Add mcp-code-review to your MCP client configuration.

Claude Code (~/.claude/claude_desktop_config.json):

{
  "mcpServers": {
    "mcp-code-review": {
      "command": "mcp-code-review",
      "env": {
        "GITHUB_TOKEN": "ghp_your_token_here",
        "GITLAB_TOKEN": "glpat_your_token_here"
      }
    }
  }
}

Cursor (.cursor/mcp.json in project root or ~/.cursor/mcp.json globally):

{
  "mcpServers": {
    "mcp-code-review": {
      "command": "mcp-code-review",
      "env": {
        "GITHUB_TOKEN": "ghp_your_token_here"
      }
    }
  }
}

Once configured, tools are available in the assistant context. Example prompts:

• "Review the PR at https://github.com/org/repo/pull/42"
• "Review the local diff against main, focus on security"
• "Review the file src/auth.py"

---

Tools

Tool	Description
review_github_pr	Fetch and review a GitHub Pull Request by URL
review_gitlab_mr	Fetch and review a GitLab Merge Request by URL
review_diff	Review local git diff against a base branch
review_file	Review a single file with static analysis
review_project	Scan an entire project: structure, deps, security
post_github_review	Post review comments on a GitHub PR
post_gitlab_review	Post review comments on a GitLab MR
save_review_report	Save a review report as a Markdown file
get_conventions	Get project conventions from config or auto-detect
list_analyzers	List available static analysis tools

All tools accept a focus parameter ("all", "security", "performance", "quality") and a locale parameter for report language.

---

Configuration

Environment Variables

Variable	Description	Default
GITHUB_TOKEN	GitHub personal access token	—
GITLAB_TOKEN	GitLab personal access token	—
GITLAB_URL	GitLab instance base URL	https://gitlab.com
REVIEW_LOCALE	Default report language (en, it, es)	en
REVIEW_OUTPUT_DIR	Directory for saved review reports	.

.codereview.yml

Place a .codereview.yml file at the root of a project to configure conventions:

language: python
framework: fastapi     # optional: django, flask, fastapi
locale: en             # en | it | es

ignore:
  - "migrations/**"
  - "tests/**"

severity_overrides:
  E501: warning        # downgrade line-length errors to warnings

custom_rules:
  max_function_lines: 50
  require_docstrings: true

If no .codereview.yml is found, the server auto-detects the language and framework from project files (pyproject.toml, requirements.txt, import patterns).

---

Multi-Language Support

Review reports can be generated in English, Italian, or Spanish. Set the language per-request via the locale parameter, or globally via the REVIEW_LOCALE environment variable or the locale key in .codereview.yml.

Supported locales:

Code	Language
en	English
it	Italian
es	Spanish

---

Claude Code Skill (Optional)

The repository includes a companion skill for Claude Code that provides a structured review process on top of the MCP tools — checklist, severity classification, report format, and multi-language support.

Install the skill

# Symlink into Claude Code skills directory
ln -s /path/to/mcp-code-review/skills/code-review-mcp ~/.claude/skills/code-review-mcp

Or copy the file manually:

mkdir -p ~/.claude/skills/code-review-mcp
cp /path/to/mcp-code-review/skills/code-review-mcp/SKILL.md ~/.claude/skills/code-review-mcp/

The skill activates automatically when you ask for a code review in Claude Code. It orchestrates the MCP tools in the right order and produces a structured markdown report.

What the skill adds

Without skill	With skill
You call tools manually	Tools called automatically in the right order
No structured process	13-category engineering checklist (see below)
Same depth for every change	Review depth scales to change size (trivial to large)
No change-type awareness	Different priorities for feature/bugfix/refactor/migration
Raw data returned	Full markdown report with severity classification
No verdict	Verdict: approved / changes requested / rejected
You decide the format	Consistent format across all reviews

Analysis Categories

The skill applies professional software engineering standards across 13 categories:

#	Category	Key checks
1	PR/MR Description	Purpose, scope, testing instructions, breaking changes
2	Architecture & Design	SOLID principles, coupling/cohesion, design patterns, separation of concerns
3	Correctness	Logic errors, edge cases, race conditions, contract violations
4	Security	OWASP top 10, injection, auth, crypto, secrets, CSRF, rate limiting
5	Performance	N+1 queries, algorithmic complexity, caching, connection pooling
6	Error Handling	Specific exceptions, resource cleanup, fail-fast, graceful degradation
7	Code Quality	Naming, DRY, nesting depth, magic numbers, abstraction level
8	Testing	Behavior tests, edge cases, minimal mocking, regression tests
9	API Design	Backward compatibility, REST conventions, versioning
10	Data Layer	Migration safety, rollback plan, schema design, Django-specific
11	Dependencies	License, maintenance, security, version pinning, necessity
12	Observability	Logging, metrics, health checks, tracing, PII protection
13	Concurrency/Async	Deadlocks, idempotency, retry safety, Celery, async/await

---

Development

Clone the repository and install dependencies with uv:

git clone https://github.com/mauriziomocci/mcp-code-review.git
cd mcp-code-review
uv sync --extra python

Run the test suite:

uv run pytest tests/ -v

Run linting:

uv run ruff check src/

---

License

MIT License. See LICENSE for details.