AI Code Agent MCP Server

A comprehensive Model Context Protocol (MCP) server that provides AI-powered code analysis capabilities using both Codex CLI and Gemini CLI. This server enables intelligent, automated code analysis with support for multiple AI agents, finding aggregation, and detailed security and performance assessments.

Features

• Dual AI Review Engines: Leverage both Codex and Gemini for comprehensive code analysis
• Combined Reviews: Aggregate findings from multiple reviewers with intelligent deduplication
• Type-Safe: Built with TypeScript and Zod for runtime validation
• Configurable: Extensive configuration options via JSON, environment variables, or code
• Production-Ready: Includes retry logic, error handling, logging, and monitoring
• Secure: Input validation, CLI sanitization, and security-first design
• Flexible: Support for multiple programming languages and review focus areas

Architecture Overview

+---------------------------------------------------------------+
|                      MCP Client (Claude)                       |
+---------------------------+-----------------------------------+
                            | MCP Protocol (stdio)
                            v
+---------------------------------------------------------------+
|                 Code Review MCP Server                         |
|  +------------------+         +------------------+            |
|  |  Codex Service   |         |  Gemini Service  |            |
|  |  (CLI Direct)    |         |  (CLI Direct)    |            |
|  +--------+---------+         +--------+---------+            |
|           |                            |                       |
|           +------------+---------------+                       |
|                        v                                       |
|           +--------------------+                               |
|           | Review Aggregator  |                               |
|           | & Deduplication    |                               |
|           +--------------------+                               |
+---------------------------------------------------------------+

Installation

From NPM (Recommended)

# Install globally
npm install -g ai-code-agent-mcp

# Or use directly with npx
npx ai-code-agent-mcp

NPM Package: ai-code-agent-mcp

From Source

git clone https://github.com/physics91/ai-code-review-mcp.git
cd ai-code-review-mcp
npm install
npm run build
npm link

Using Docker

docker pull code-review-mcp:latest
docker run -v ./config.json:/config.json code-review-mcp

Prerequisites

• Node.js 20.0.0 or higher
• Gemini CLI installed and configured (for Gemini reviews)
• Codex CLI installed and configured (for Codex reviews)

Installing CLIs

Codex CLI

# Install from npm
npm install -g @anthropic-ai/codex

# Verify installation
codex --version

Gemini CLI

# Example installation (adjust based on your system)
npm install -g @google/gemini-cli
# or
brew install gemini-cli

CLI Path Auto-Detection

The server automatically detects CLI paths based on your platform and environment. You don't need to specify the exact path in most cases!

Priority Order

1. Environment Variables (highest priority)

   • CODEX_CLI_PATH - Custom Codex CLI path
   • GEMINI_CLI_PATH - Custom Gemini CLI path

2. Config File - Explicit path in config.json

3. Platform-Specific Paths (auto-detected)

   • macOS / Linux:

     • /usr/local/bin/{cli}
     • /usr/bin/{cli}
     • /opt/{cli}/bin/{cli}
     • ~/.local/bin/{cli}
     • /opt/homebrew/bin/{cli} (macOS Homebrew)

   • Windows:

     • %APPDATA%\npm\{cli}.cmd
     • C:\Program Files\{cli}\{cli}.exe
     • C:\Program Files\Google\Gemini\gemini.exe (Gemini only)

4. System PATH - which (Unix) or where (Windows) command

5. Fallback - Assumes CLI is in PATH

Configuration Options

Option 1: Auto-Detection (Recommended)

{
  "codex": {
    "cliPath": "auto"  // Automatically detects CLI path
  },
  "gemini": {
    "cliPath": "auto"  // Automatically detects CLI path
  }
}

Option 2: Environment Variables

# Set custom CLI paths
export CODEX_CLI_PATH="/custom/path/codex"
export GEMINI_CLI_PATH="/opt/google/gemini/gemini"

Option 3: Explicit Configuration

{
  "codex": {
    "cliPath": "/usr/local/bin/codex"
  },
  "gemini": {
    "cliPath": "/opt/gemini/bin/gemini"
  }
}

Option 4: Default Command Name

{
  "codex": {
    "cliPath": "codex"  // Uses 'codex' from PATH
  },
  "gemini": {
    "cliPath": "gemini"  // Uses 'gemini' from PATH
  }
}

Detection Logs

The server logs the detected CLI paths on startup:

[INFO] Codex CLI path detected {
  path: "/usr/local/bin/codex",
  source: "detected",
  exists: true,
  platform: "darwin"
}

[INFO] Gemini CLI path detected {
  path: "/opt/homebrew/bin/gemini",
  source: "detected",
  exists: true,
  platform: "darwin"
}

Detection sources:

• env - From environment variable
• config - From configuration file
• detected - Auto-detected from platform-specific paths
• which - Found using which/where command
• default - Fallback to command name (may fail if not in PATH)

Configuration

Claude Desktop Integration

Add to your Claude Desktop configuration (claude_desktop_config.json):

{
  "mcpServers": {
    "code-review": {
      "command": "node",
      "args": ["/path/to/code-review-mcp/dist/index.js"],
      "env": {
        "CODE_REVIEW_MCP_LOG_LEVEL": "info",
        "CODEX_ENABLED": "true",
        "GEMINI_ENABLED": "true"
      }
    }
  }
}

Note: CLI paths are auto-detected by default. Only set CODEX_CLI_PATH or GEMINI_CLI_PATH if you need to override the detection.

Configuration File

Create a config.json in your project or use the default configuration:

{
  "server": {
    "name": "code-review-mcp",
    "version": "1.2.0",
    "logLevel": "info"
  },
  "codex": {
    "enabled": true,
    "cliPath": "auto",
    "timeout": 60000,
    "retryAttempts": 3,
    "model": "gpt-5",
    "search": true,
    "reasoningEffort": "high",
    "args": []
  },
  "gemini": {
    "enabled": true,
    "cliPath": "auto",
    "timeout": 60000,
    "retryAttempts": 3,
    "model": "gemini-3-pro-preview",
    "args": []
  },
  "review": {
    "maxCodeLength": 50000,
    "deduplication": {
      "enabled": true,
      "similarityThreshold": 0.8
    }
  }
}

Note: "cliPath": "auto" enables automatic CLI path detection. You can also specify:

• "auto" - Auto-detect (recommended)
• "codex" / "gemini" - Use command from PATH
• Absolute path - Explicit path to CLI binary

Environment Variables

Create a .env file:

# Copy example configuration
cp .env.example .env

# Edit with your settings
CODEX_CLI_PATH=codex
CODEX_MODEL=claude-opus-4
GEMINI_CLI_PATH=/usr/local/bin/gemini
GEMINI_MODEL=gemini-pro
CODEX_ENABLED=true
GEMINI_ENABLED=true

Usage

Available Tools

The MCP server exposes the following tools:

1. analyze_code_with_codex

Perform code analysis using Codex AI.

Input Parameters:

{
  code: string;                    // Source code to review (max 50KB)
  language?: string;               // Programming language (auto-detect if omitted)
  context?: {
    fileName?: string;             // File name for context
    projectType?: string;          // e.g., "web", "backend", "mobile"
    reviewFocus?: Array<           // What to focus on
      'security' | 'performance' | 'style' | 'bugs' | 'all'
    >;
  };
  options?: {
    timeout?: number;              // Timeout in ms (default: 60000)
    includeExplanations?: boolean; // Include detailed explanations
    severity?: 'all' | 'high' | 'medium'; // Filter by severity
  };
}

Example Usage in Claude:

Please analyze this code using Codex:

[Call: analyze_code_with_codex]
{
  "code": "function calculateTotal(items) { let total = 0; for(let i=0; i<=items.length; i++) { total += items[i].price; } return total; }",
  "language": "javascript",
  "context": {
    "fileName": "cart.js",
    "reviewFocus": ["bugs", "security"]
  }
}

Output:

{
  "success": true,
  "reviewId": "uuid",
  "timestamp": "2025-01-17T10:30:00.000Z",
  "source": "codex",
  "summary": {
    "totalFindings": 3,
    "critical": 1,
    "high": 1,
    "medium": 1,
    "low": 0
  },
  "findings": [
    {
      "type": "bug",
      "severity": "critical",
      "line": 1,
      "title": "Off-by-one error in loop",
      "description": "Loop condition uses <= instead of <, causing array index out of bounds",
      "suggestion": "Change i<=items.length to i<items.length",
      "code": "for(let i=0; i<items.length; i++)"
    }
  ],
  "overallAssessment": "Code has critical bugs that need immediate fixing",
  "recommendations": [
    "Add input validation for items parameter",
    "Consider using Array.reduce() for cleaner code"
  ]
}

2. analyze_code_with_gemini

Perform code analysis using Gemini CLI.

Input Parameters: Same as analyze_code_with_codex

Example Usage:

Analyze this Python code with Gemini:

[Call: analyze_code_with_gemini]
{
  "code": "def process_data(data):\n    result = []\n    for item in data:\n        result.append(item * 2)\n    return result",
  "language": "python",
  "context": {
    "reviewFocus": ["performance", "style"]
  }
}

3. analyze_code_combined

Perform code analysis using both Codex and Gemini, then aggregate results.

Input Parameters:

{
  code: string;
  language?: string;
  context?: {
    fileName?: string;
    projectType?: string;
    reviewFocus?: Array<'security' | 'performance' | 'style' | 'bugs' | 'all'>;
  };
  options?: {
    timeout?: number;              // Timeout for entire operation
    includeExplanations?: boolean;
    severity?: 'all' | 'high' | 'medium';
    parallelExecution?: boolean;   // Run both analyzers in parallel
    includeIndividualAnalyses?: boolean; // Include separate analyses in output
  };
}

Example Usage:

Please perform a comprehensive analysis using both Codex and Gemini:

[Call: analyze_code_combined]
{
  "code": "class UserAuth { login(user, pass) { if(user && pass) { return db.query('SELECT * FROM users WHERE username=' + user); } } }",
  "language": "javascript",
  "context": {
    "fileName": "auth.js",
    "reviewFocus": ["security", "bugs"]
  },
  "options": {
    "parallelExecution": true,
    "includeIndividualAnalyses": false
  }
}

Output:

{
  "success": true,
  "reviewId": "uuid",
  "timestamp": "2025-01-17T10:30:00.000Z",
  "source": "combined",
  "summary": {
    "totalFindings": 5,
    "critical": 2,
    "high": 2,
    "medium": 1,
    "low": 0,
    "consensus": 85
  },
  "findings": [
    {
      "type": "security",
      "severity": "critical",
      "line": 4,
      "title": "SQL Injection vulnerability",
      "description": "User input directly concatenated into SQL query",
      "suggestion": "Use parameterized queries or ORM",
      "sources": ["codex", "gemini"],
      "confidence": "high"
    }
  ],
  "overallAssessment": "Combined analysis from 2 analyzers: Found 2 critical issues that require immediate attention.",
  "metadata": {
    "analysisDuration": 4523,
    "codexDuration": 2341,
    "geminiDuration": 2182
  }
}

4. get_analysis_status

Check the status of an async analysis operation (for future async support).

Input Parameters:

{
  analysisId: string; // UUID of the analysis
}

Advanced Usage

Custom Review Focus

{
  "code": "...",
  "context": {
    "reviewFocus": ["security", "performance"]
  }
}

Focus areas:

• security: SQL injection, XSS, authentication issues
• performance: Inefficient algorithms, memory leaks
• style: Code formatting, naming conventions
• bugs: Logic errors, edge cases
• all: Comprehensive review (default)

Severity Filtering

{
  "code": "...",
  "options": {
    "severity": "high" // Only show critical and high severity issues
  }
}

Parallel Execution

For faster combined analyses:

{
  "code": "...",
  "options": {
    "parallelExecution": true // Run Codex and Gemini concurrently
  }
}

Configuration Reference

Server Configuration

| Option | Type | Default | Description | |--------|------|---------|-------------| | server.name | string | "ai-code-agent-mcp" | Server name | | server.version | string | "1.2.0" | Server version | | server.logLevel | string | "info" | Log level (debug/info/warn/error) |

Codex Configuration

| Option | Type | Default | Description | |--------|------|---------|-------------| | codex.enabled | boolean | true | Enable Codex analysis | | codex.timeout | number | 60000 | Timeout in milliseconds | | codex.retryAttempts | number | 3 | Number of retry attempts | | codex.model | string | null | Codex model override |

Gemini Configuration

| Option | Type | Default | Description | |--------|------|---------|-------------| | gemini.enabled | boolean | true | Enable Gemini analysis | | gemini.cliPath | string | "/usr/local/bin/gemini" | Path to Gemini CLI | | gemini.timeout | number | 60000 | Timeout in milliseconds | | gemini.retryAttempts | number | 3 | Number of retry attempts | | gemini.model | string | null | Gemini model override |

Analysis Configuration

| Option | Type | Default | Description | |--------|------|---------|-------------| | analysis.maxCodeLength | number | 50000 | Maximum code length in bytes | | analysis.deduplication.enabled | boolean | true | Enable finding deduplication | | analysis.deduplication.similarityThreshold | number | 0.8 | Similarity threshold (0-1) |

Error Handling

The server includes comprehensive error handling:

// Errors are returned in a structured format
{
  "success": false,
  "error": {
    "type": "TIMEOUT_ERROR",
    "message": "Review timed out after 60000ms",
    "code": "ERR_TIMEOUT",
    "details": { ... }
  }
}

Common error types:

• VALIDATION_ERROR: Invalid input parameters
• TIMEOUT_ERROR: Operation timed out
• CLI_EXECUTION_ERROR: CLI execution failed
• PARSE_ERROR: Failed to parse CLI output
• CONFIGURATION_ERROR: Invalid configuration

Security Considerations

1. Input Validation: All inputs are validated using Zod schemas
2. CLI Safety: CLI paths are whitelisted, no shell injection possible
3. Code Length Limits: Prevents DoS via large payloads
4. Sanitization: Sensitive data is redacted from logs
5. Retry Limits: Prevents infinite retry loops

Performance

Performance targets:

• Single analysis: <5s (typical), <30s (maximum)
• Combined analysis: <8s (typical), <60s (maximum)
• Memory usage: <200MB (active), <50MB (idle)
• Concurrent reviews: 10 (default), 50 (maximum)

Logging

Structured JSON logs with Pino:

{
  "level": "info",
  "timestamp": "2025-01-17T10:30:00.000Z",
  "msg": "Analysis completed",
  "analysisId": "uuid",
  "source": "codex",
  "duration": 4532,
  "findings": 12
}

Set log level via environment:

CODE_REVIEW_MCP_LOG_LEVEL=debug

Development

Setup

git clone https://github.com/physics91/ai-code-agent-mcp.git
cd ai-code-agent-mcp
npm install

Development Mode

npm run dev

Build

npm run build

Testing

# Run tests
npm test

# Run tests with coverage
npm run test:coverage

# Run tests with UI
npm run test:ui

Type Checking

npm run typecheck

Linting

npm run lint
npm run lint:fix

Troubleshooting

Codex MCP tool not found

Solution: Ensure the Codex MCP tool is properly registered in your MCP client.

Gemini CLI execution fails

Solution:

1. Verify Gemini CLI is installed: gemini --version
2. Check CLI path in configuration
3. Ensure executable permissions: chmod +x /path/to/gemini

Analysis timeout

Solution:

1. Increase timeout in configuration
2. Reduce code length
3. Check system resources

High memory usage

Solution:

1. Disable cache if not needed
2. Reduce maxConcurrent setting
3. Check logs for memory leaks

Documentation

For detailed documentation, please refer to:

Core Documentation

• Implementation Guide - Step-by-step implementation guide
• Usage Examples - Real-world usage examples
• Project Summary - Project overview and objectives

Reference

• Architecture - System architecture and design
• Specifications - Technical specifications

Development History

• Migration Documentation - CLI migration guides
• Development Notes - Implementation status and fixes

Contributing

Contributions are welcome! Please read our contributing guidelines and code of conduct.

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests
5. Submit a pull request

License

MIT License - see LICENSE file for details

Support

• GitHub Issues: https://github.com/physics91/ai-code-agent-mcp/issues
• Documentation: https://github.com/physics91/ai-code-agent-mcp/wiki
• Discussions: https://github.com/physics91/ai-code-agent-mcp/discussions

Roadmap

Short-term (3-6 months)

• [ ] Support for additional AI analyzers
• [ ] Custom analysis templates
• [ ] Analysis history and analytics
• [ ] Webhook notifications
• [ ] Multi-file project analysis
• [ ] AI debate/discussion features

Long-term (6-12 months)

• [ ] ML-based finding prioritization
• [ ] CI/CD integrations
• [ ] Analysis collaboration features
• [ ] Plugin system
• [ ] Web dashboard

Acknowledgments

• Built with @modelcontextprotocol/sdk
• Powered by Codex and Gemini AI
• Inspired by the MCP community

Version: 1.2.0 Last Updated: 2025-12-15 Status: Production Ready