Claude_MCP_Bridge

Connect Claude Code with multiple AI models (Gemini, Grok, ChatGPT, DeepSeek) through a unified Model Context Protocol interface

Overview

The Claude_MCP_Bridge enables Claude Code to collaborate with multiple AI models simultaneously. Compare responses, facilitate AI debates, and leverage collective intelligence for complex problem-solving directly from your terminal.

Key Features

Multi-Model Integration: Connect with Gemini, Grok, ChatGPT, and DeepSeek through one interface
AI Collaboration Tools: Enable debates, consensus-building, and collaborative problem-solving
Specialized Workflows: Code review, debugging, architecture design, and brainstorming
Flexible Configuration: Use only the AI models you have access to
Unified Interface: Consistent tool patterns across all AI providers

Quick Start

Prerequisites

Python 3.7+ - Download here
Claude Code CLI - Install with: npm install -g @anthropic-ai/claude-code
AI API Keys - At least one from supported providers (see below)

Installation

Automated Setup

# Clone and run setup
git clone <repository-url>
cd multi-ai-mcp
chmod +x setup.sh
./setup.sh

The setup script will:

Install Python dependencies
Configure API keys interactively
Register the MCP server with Claude Code

Supported AI Providers

Gemini - Free API key from Google AI Studio
Grok - API key from xAI Console
OpenAI - API key from OpenAI Platform
DeepSeek - API key from DeepSeek Platform

Usage

Starting Claude Code

claude

Verify Connection

/mcp
# Should show: Claude_MCP_Bridge connected

Tool Categories

Status and Configuration

server_status - Check available AIs and configuration

mcp__Claude_MCP_Bridge__server_status

Individual AI Tools

Each configured AI provides these tools:

ask_{ai_name} - Direct questions

mcp__Claude_MCP_Bridge__ask_gemini "Explain containerization best practices"

{ai_name}_code_review - Code analysis with focus areas

mcp__Claude_MCP_Bridge__gemini_code_review

Focus options: general, security, performance, readability

{ai_name}_think_deep - Extended reasoning and analysis

mcp__Claude_MCP_Bridge__grok_think_deep

{ai_name}_brainstorm - Creative problem-solving

mcp__Claude_MCP_Bridge__openai_brainstorm

{ai_name}_debug - Debugging assistance

mcp__Claude_MCP_Bridge__deepseek_debug

{ai_name}_architecture - System design recommendations

mcp__Claude_MCP_Bridge__gemini_architecture

Multi-AI Collaboration Tools

ask_all_ais - Compare responses from all available AIs

mcp__Claude_MCP_Bridge__ask_all_ais "Database selection for real-time analytics"

ai_debate - Facilitate debate between two AIs

mcp__Claude_MCP_Bridge__ai_debate "Monolith vs microservices for startups" gemini grok

collaborative_solve - Multi-AI problem-solving session

mcp__Claude_MCP_Bridge__collaborative_solve "Design a distributed cache system"

ai_consensus - Get unified recommendations

mcp__Claude_MCP_Bridge__ai_consensus "Best frontend framework for enterprise apps"

Configuration

Configuration File Location

~/.Claude_MCP_Bridge/credentials.json

Configuration Structure

{
  "gemini": {
    "api_key": "your-api-key",
    "model": "gemini-2.5-flash",
    "enabled": true
  },
  "grok": {
    "api_key": "your-api-key",
    "model": "grok-4",
    "base_url": "https://api.x.ai/v1",
    "enabled": true
  },
  "openai": {
    "api_key": "your-api-key",
    "model": "gpt-4o",
    "enabled": true
  },
  "deepseek": {
    "api_key": "your-api-key",
    "model": "deepseek-chat",
    "base_url": "https://api.deepseek.com",
    "enabled": true
  }
}

Available Models

Gemini Models

gemini-2.5-pro - Most capable model with thinking mode
gemini-2.5-flash - Fast and efficient (default)
gemini-2.5-flash-lite - Lightweight variant
gemini-2.5-flash-preview-05-20 - Preview release
gemini-2.5-pro-preview-06-05 - Pro preview release
gemini-2.0-flash - Previous generation
gemini-2.0-flash-lite - Previous gen lightweight
gemma-3n-e4b-it - Open model variant

Grok Models

grok-4 - Latest model (default)
grok-4-latest - Rolling latest version
grok-4-fast-reasoning - Optimized for reasoning tasks
grok-4-fast-non-reasoning - Optimized for general tasks
grok-3 - Previous generation
grok-3-mini - Lightweight version
grok-code-fast-1 - Specialized for coding
grok-2-image-1212 - Image generation model

OpenAI Models

gpt-5 - Flagship model with unified reasoning (default)
gpt-5-codex - GPT-5 for agentic coding
gpt-4.1 - Latest GPT-4.1 series
gpt-4.1-mini - Smaller GPT-4.1 variant
gpt-4.1-nano - Smallest GPT-4.1 variant
gpt-4.5 - Previous generation (deprecating July 2025)
gpt-4o - Optimized GPT-4
gpt-4o-mini - Smaller optimized variant
o4-mini - O4 reasoning model
o3-mini - O3 reasoning model
gpt-oss-120b - Open weights 120B model
gpt-oss-20b - Open weights 20B model

DeepSeek Models

deepseek-chat - V3.1-Terminus non-thinking mode (default)
deepseek-reasoner - V3.1-Terminus R1 thinking mode

Architecture

System Design

┌─────────────────┐
│   Claude Code   │
└────────┬────────┘
         │ MCP Protocol (JSON-RPC 2.0)
┌────────▼─────────────────────────┐
│      Claude_MCP_Bridge           │
│ ┌──────────────────────────────┐ │
│ │   Unified Tool Interface     │ │
│ └──┬─────┬─────┬─────┬───────-─┘ │
│    │     │     │     │           │
│ ┌──▼──┐┌─▼──┐ ┌─▼──┐ ┌─▼──────┐  │
│ │Gemini││Grok││OAI │ │DeepSeek│  │
│ └──────┘└────┘└────┘ └────────┘  │
└──────────────────────────────────┘

Protocol Details

Protocol: JSON-RPC 2.0 over stdin/stdout
MCP Version: 2024-11-05
Response Format: Unified across all AI providers
Error Handling: Graceful degradation when AIs unavailable

Examples

Comparative Analysis

# Compare AI perspectives on a technical decision
mcp__Claude_MCP_Bridge__ask_all_ais "PostgreSQL vs MongoDB for time-series data"

Code Review Pipeline

# Sequential reviews focusing on different aspects
mcp__Claude_MCP_Bridge__gemini_code_review "$(cat app.py)" security
mcp__Claude_MCP_Bridge__grok_code_review "$(cat app.py)" performance
mcp__Claude_MCP_Bridge__deepseek_code_review "$(cat app.py)" maintainability

Architecture Planning

# Collaborative architecture design
mcp__Claude_MCP_Bridge__collaborative_solve "Design event-driven microservices for e-commerce"

Technical Debate

# AI debate on architectural choices
mcp__Claude_MCP_Bridge__ai_debate "REST vs GraphQL for mobile applications" openai deepseek

Troubleshooting

MCP Connection Issues

# Remove and re-add the server
claude mcp remove Claude_MCP_Bridge
claude mcp add --scope user Claude_MCP_Bridge python3 ~/.Claude_MCP_Bridge/server.py

Verify Installation

# Check Python dependencies
pip3 list | grep -E "google-genai|openai|python-dotenv"

# Test server directly
python3 ~/.Claude_MCP_Bridge/server.py

Update API Keys

# Edit configuration
nano ~/.Claude_MCP_Bridge/credentials.json

# Or use environment variables
export GEMINI_API_KEY="new-key"
export OPENAI_API_KEY="new-key"

Common Issues

Issue: AI not responding

Solution: Check API key validity and rate limits

Issue: Model not found

Solution: Verify model name in configuration matches available models

Issue: MCP server not connecting

Solution: Ensure Claude Code CLI is updated to latest version

API Limits and Costs

Rate Limits

| Provider | Free Tier | Paid Tier | |----------|-----------|-----------| | Gemini | 15 RPM, 1500/day | 360 RPM | | Grok | N/A | 60 RPM | | OpenAI | N/A | 500 RPM (GPT-4o) | | DeepSeek | 50 RPM | 500 RPM |

Token Limits

| Provider | Context Window | Max Output | Special Features | |----------|---------------|------------|------------------| | Gemini 2.5 | Up to 2M tokens | 8192 tokens | Thinking mode (Pro), Live API, Image generation | | Grok-4 | 256k tokens | 8192 tokens | Dynamic reasoning, Live search via API | | GPT-5 | Up to 1M tokens | 8192 tokens | Unified reasoning, Codex integration | | DeepSeek V3.1 | 64k tokens | 8192 tokens | R1 thinking mode, Context caching |