Skip to main content
Glama
MCP Code Sanitizer

MCP Code Sanitizer

by notasandy

mcp-code-sanitizer

Strict AI-powered code reviewer for Claude Desktop, Cursor, VS Code, and Claude Code CLI. Finds bugs, vulnerabilities, and security issues — powered by Groq (free API).

Python PyPI FastMCP Groq License

Claude / Cursor / VS Code  ──MCP──►  code-sanitizer  ──REST──►  Groq API
                                        (server.py)              (llama-3.3-70b)

demo


Features

Tool

What it does

analyze_code

Strict review — bugs, security issues, score 0–100

compare_code

Compares two versions, detects regressions, recommends merge/request_changes

explain_code

Step-by-step explanation for junior / middle / senior audience

generate_tests

Generates pytest / jest / go test — happy path, edge cases, security

analyze_file

Analyzes a whole file from disk with parallel chunking

generate_report

Builds an HTML report from any analysis result

cache_info

Cache statistics and clearing

Example output

{
  "summary": "Critical SQL injection and secret exposed in logs",
  "score": 23,
  "issues": [
    {
      "severity": "critical",
      "line": 2,
      "title": "SQL Injection",
      "description": "f-string directly interpolates user_id into query",
      "fix": "cursor.execute('SELECT * FROM users WHERE id = %s', (user_id,))"
    }
  ],
  "warnings": [{"title": "No exception handling", "description": "..."}],
  "suggestions": ["Consider using an ORM instead of raw SQL"]
}

Related MCP server: Claude Code Review MCP

Installation

Prerequisite: Get a free Groq API key at console.groq.com/keys — no credit card required.

Claude Code CLI

claude mcp add code-sanitizer -e GROQ_API_KEY=gsk_your_key -- uvx mcp-code-sanitizer

Claude Desktop

OS

Config file

macOS

~/Library/Application Support/Claude/claude_desktop_config.json

Windows

%APPDATA%\Claude\claude_desktop_config.json

Linux

~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "code-sanitizer": {
      "command": "uvx",
      "args": ["mcp-code-sanitizer"],
      "env": {
        "GROQ_API_KEY": "gsk_your_key_here"
      }
    }
  }
}

Cursor

Create .cursor/mcp.json in your project (or ~/.cursor/mcp.json globally):

{
  "mcpServers": {
    "code-sanitizer": {
      "command": "uvx",
      "args": ["mcp-code-sanitizer"],
      "env": {
        "GROQ_API_KEY": "gsk_your_key_here"
      }
    }
  }
}

VS Code

Requires VS Code 1.99+ with GitHub Copilot. Create .vscode/mcp.json in your project:

{
  "servers": {
    "code-sanitizer": {
      "command": "uvx",
      "args": ["mcp-code-sanitizer"],
      "env": {
        "GROQ_API_KEY": "gsk_your_key_here"
      }
    }
  }
}

Or add globally via Ctrl+Shift+P → "MCP: Add Server".

Don't have uvx? Install it with pip install uv, then use the commands above.


Manual install (alternative)

If you prefer cloning the repo:

git clone https://github.com/notasandy/mcp-code-sanitizer
cd mcp-code-sanitizer
pip install -r requirements.txt
cp .env.example .env   # add your GROQ_API_KEY
python server.py

Then point the client config to:

{
  "command": "python",
  "args": ["/full/path/to/server.py"],
  "env": { "GROQ_API_KEY": "gsk_your_key_here" }
}

GitHub Action — automatic PR review

Add AI code review to any repository in 5 lines. The action posts a structured comment on every PR with score, issues, and fix suggestions.

# .github/workflows/ai-review.yml
name: AI Code Review
on:
  pull_request:
    types: [opened, synchronize]

permissions:
  contents: read
  pull-requests: write

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: notasandy/mcp-code-sanitizer@v1
        with:
          groq_api_key: ${{ secrets.GROQ_API_KEY }}

Add GROQ_API_KEY to your repository secrets → Settings → Secrets → Actions.

The action automatically:

  • Reviews only changed files (up to 10 per PR)

  • Posts a score and structured issue list as a PR comment

  • Fails the check if critical issues are found


Usage in chat

After connecting, just write naturally:

Review this code for vulnerabilities:

def get_user(user_id):
    query = f"SELECT * FROM users WHERE id = {user_id}"
    return db.execute(query)

Or call tools explicitly:

analyze_file /path/to/my_script.py
generate_tests for this function: ...
compare_code — before vs after refactor, did it get better?
generate_report and save to /tmp/report.html

Architecture

mcp-code-sanitizer/
├── server.py          # FastMCP entry point
├── config.py          # Constants — keys, limits, extension map
├── groq_client.py     # Async Groq client with auto-retry on 429
├── cache.py           # In-memory LRU cache with TTL
├── prompts.py         # System prompts for all tools
└── tools/
    ├── analyze.py     # analyze_code
    ├── compare.py     # compare_code
    ├── explain.py     # explain_code
    ├── tests.py       # generate_tests
    ├── file_tool.py   # analyze_file — chunking + parallel analysis
    ├── cache_tool.py  # cache_info
    └── report.py      # generate_report — HTML output

Configuration

All settings via .env or environment variables:

Variable

Default

Description

GROQ_API_KEY

—

Required. Get at console.groq.com

GROQ_MODEL

llama-3.3-70b-versatile

Groq model to use

CACHE_TTL

3600

Cache TTL in seconds

CACHE_MAX

200

Max cached entries

Available Groq models

Model

Speed

Quality

llama-3.3-70b-versatile

Fast

Best (default)

llama-3.1-8b-instant

Fastest

Good

mixtral-8x7b-32768

Fast

Great


Contributing

PRs and Issues are welcome. Most wanted:

  • Support for other LLM providers (OpenAI, Anthropic)

  • New tools: dependency audit, complexity score, docstring generator

  • Prompt improvements and new language support


License

MIT — do whatever you want. A star would be appreciated.


Available Tools

7 tools
analyze_codeC

Strict analysis of a code fragment using Groq LLM.

Inputs: code (required), language, context. Returns: result

analyze_fileA

Analyzes a whole code file from disk. Automatically detects language by file extension. Large files are split into chunks and analyzed in parallel.

Inputs: file_path (required), language, context. Returns: result

cache_infoA

Shows cache statistics or clears the cache Args: clear: True - clears the cache, False - shows statistics.

Returns: JSON with cache stats or clear result.

Inputs: clear. Returns: result

compare_codeA

Compares two versions of code and evaluates whether the change is an improvement.

Performs a structured diff analysis: identifies what improved, what regressed, and what changed neutrally. Returns a merge recommendation based on the findings. Useful for code review, refactoring validation, and AI-generated code verification.

Inputs: code_before (required), code_after (required), language, context. Returns: result

explain_codeA

Explains what code does - step by step and clearly. Args: code: Code to explain. language: Programming language. audience: Target audience level - junior, middle, or senior. Returns: JSON with step-by-step explanation, key concepts, and gotchas.

Inputs: code (required), language, audience. Returns: result

generate_reportA

Generates a beautiful HTML report from analyze_code or analyze_file results. Args: analysis_json: JSON string from analyze_code or analyze_file. output_path: Path to save the HTML file (optional). source_name: File/fragment name for the report title. Returns: JSON with fields: html, saved_to, length.

Inputs: analysis_json (required), output_path, source_name. Returns: result

generate_testsA

Generates tests for the provided code. Args: code: Code to generate tests for. language: Programming language. framework: Test framework (optional - pytest, jest, unittest, etc.). Returns: JSON with test cases, runnable test code, and coverage estimate.

Inputs: code (required), language, framework. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 7 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose: code fragment analysis, whole file analysis, cache management, code comparison, code explanation, report generation, and test generation. No overlap in functionality.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern using snake_case (e.g., analyze_code, generate_tests). No inconsistencies in style.

    Tool Count5/5

    7 tools is a well-scoped set for code analysis and sanitization tasks, covering the core operations without being overwhelming or too sparse.

    Completeness4/5

    The tool surface covers analysis, comparison, explanation, test generation, and reporting. A minor gap is the lack of code transformation or refactoring tools, but the set is reasonable for the stated domain.

    Maintenance

    ActivityInactive
    ResponsivenessLimited data

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      A server that enables interaction with PostgreSQL, MySQL, MariaDB, or SQLite databases through Claude Desktop using natural language queries.
      1
      -
    • A
      license
      A
      quality
      D
      maintenance
      Enables comprehensive security scanning of code projects, detecting vulnerabilities in dependencies, code patterns (XSS, eval, etc.), and exposed secrets, with detailed reports in Spanish prioritized by severity.
      4
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      AI-powered code review tool that detects AI-generated code defects invisible to traditional linters — hallucinated packages, deprecated APIs, cross-file contradictions, hidden security anti-patterns, and over-engineering. Works as a standalone CLI, GitHub Action, or MCP server. Supports TypeScript, Python, Java, Go, and Kotlin. Free for individuals, no API key required.
      4
      41
      Business Source 1.1