Gemini CLI MCP

This is an enterprise-grade MCP server that connects Google’s Gemini CLI with MCP-compatible clients like Claude Code, Claude Desktop, VS Code, Cursor AI, and more.

It’s integrated with the popular OpenRouter API. So you and your development teams can access over 400 AI models from leading AI providers, including OpenAI, Anthropic, Meta, and Google, through a unified interface.

Features

• 🚀 33 Specialized MCP Tools – Toolset for multi-AI integration across six tool categories, including core Gemini tools, system analysis, conversation management, and specialized code review capabilities.
• 🤖 400+ AI Models Access – Access to models from OpenAI, Anthropic, Meta, Google, and 20+ additional providers with cost management and usage tracking.
• 🏗️ Enterprise Architecture – Modular design with 83 Python files organized across specialized modules, including core server layers, configuration management, integration modules, and comprehensive security frameworks.
• 💾 Stateful Conversation History – Redis-backed conversation storage with cross-platform support, automatic context building, and configurable expiration policies for persistent multi-turn interactions.
• ⚡ Dynamic Token Management – Tool-specific limits from 100K-800K characters with model-aware scaling and automatic optimization based on selected AI model capabilities.
• 🔄 Multi-AI Collaboration – Purpose-built tools for plan evaluation, code review, cross-platform collaboration, and structured AI debates with sequential, debate, and validation modes.
• 📁 @filename Syntax Support – Direct file reading capabilities with intelligent large file handling strategies across 32 tools, supporting directories, wildcards, and mixed content processing.
• 🛡️ Enterprise Security – Security framework with 22 critical security fixes, multi-layer defense systems, and real-time protection against common attack vectors.
• 📊 Production Monitoring – OpenTelemetry tracing and Prometheus metrics integration with performance tracking, cache effectiveness monitoring, and resource utilization analysis.
• ⚡ High Performance – Async architecture with lock-free cache operations, process pooling, and automatic model fallback systems supporting enterprise-scale concurrent request processing.

Use Cases

• Multi-AI Development Workflows – Development teams can orchestrate complex workflows where Claude Code generates implementation plans, Gemini AI evaluates architectural decisions, and OpenRouter models provide specialized analysis.
• Enterprise Codebase Analysis – Large development organizations can analyze massive codebases spanning multiple directories and technologies using the enhanced file processing capabilities. The system handles enterprise-scale projects with intelligent token management, supporting analysis of complete system architectures, dependency mapping, and technical debt assessment across thousands of files.
• Cross-Platform AI Collaboration – Technical teams can leverage the AI collaboration tools to facilitate structured debates between different AI models on architectural decisions, conduct multi-model validation of critical system designs, and implement sequential analysis pipelines where each AI contributes specialized expertise to complex technical challenges.
• Production Code Review Automation – Development teams can integrate the specialized code review tools into their continuous integration pipelines, automatically analyzing git diffs, extracting structured data from codebases using JSON schemas, and maintaining consistent code quality standards across distributed development teams through automated multi-AI review processes.

FAQs

Q: How does the automatic model fallback system work?
A: When quota limits are exceeded on premium models like gemini-2.5-pro, the system automatically retries the request using gemini-2.5-flash without user intervention. This ensures continuous operation during high-usage periods and maintains service availability for development teams.

Q: Can I use the server without OpenRouter integration?
A: Yes, the server provides full Gemini CLI functionality without requiring OpenRouter configuration. The OpenRouter integration is optional and adds access to 400+ additional AI models. All core features including the 33 specialized tools work independently with just Gemini CLI configured.

Q: What security measures are implemented for enterprise deployments?
A: The server includes 22 critical security fixes with multi-layer defense against command injection, path traversal, XSS, and prompt injection attacks. Security features include session-based rate limiting, template integrity validation, enhanced credential sanitization, JSON-RPC validation, and subprocess injection protection with compiled regex patterns for real-time threat detection.

Q: How do I handle large codebases that exceed token limits?
A: The system provides intelligent large file handling strategies including automatic chunking, summarization, and full processing modes. Use gemini_summarize_files for maximum capacity (800K characters), configure custom limits through environment variables, or leverage the file_handling_strategy parameter in OpenRouter tools for automatic optimization.

Q: What monitoring and performance metrics are available?
A: The gemini_metrics tool provides comprehensive performance data including command execution statistics, cache hit rates, error classifications, model usage patterns, memory utilization, and OpenRouter cost tracking. Enterprise monitoring through OpenTelemetry and Prometheus offers distributed tracing and detailed resource analysis.

Q: How does conversation history work across different AI models?
A: The conversation management system maintains stateful context across Gemini CLI and OpenRouter models with Redis-backed storage. Conversations automatically handle context building, token limit management, and cross-platform compatibility, allowing seamless transitions between different AI providers within the same conversation thread.

Q: What are the performance characteristics for concurrent usage?
A: The async architecture supports 1,000-10,000+ concurrent requests with 10-100x improvement over traditional implementations. Memory-efficient single-threaded design with non-blocking I/O operations provides average response times of 2-10 seconds for medium operations and 10-60 seconds for complex analysis tasks.