🤖 Second Opinion MCP

🎯 Get instant second opinions from 17 AI platforms and 800,000+ models

OpenAI • Gemini • Grok • Claude • HuggingFace • DeepSeek • OpenRouter • Mistral • Together AI • Cohere • Groq • Perplexity • Replicate • AI21 Labs • Stability AI • Fireworks AI • Anyscale

🚀 What it does

This MCP server allows Claude to consult other AI models for different perspectives on:

• Coding problems - Compare approaches across models

• Creative writing - Get diverse style feedback

• Problem solving - Validate logic and reasoning

• Cross-model analysis - See how different AIs tackle the same task

• Group discussions - Host AI debates with multiple models

• Custom model access - Use any HuggingFace model via Inference API

✨ Version 5.0 Features & Improvements

🎭 NEW: AI Personality System

• 5 Distinct Personalities: honest, freind, coach, wise, creative

• Intelligent Model Matching: Each personality uses models best suited for their character

• Always Available: Works with any configured AI provider

🧠 NEW: Intelligent Model Selection

• Quality-Based Ranking: 34+ models ranked by capability (Grok-4 → Gemini Pro → GPT-4.1)

• Smart Defaults: Automatically selects the best available model

• Personality Optimization: Different models for different personality types

🏗️ NEW: Modular Architecture

• 5 Clean Files: Replaced 51k+ token monolith with maintainable modules

• Professional Structure: client_manager.py, ai_providers.py, conversation_manager.py, mcp_server.py, main.py

• JSON Configuration: Easy model priority updates via model_priority.json

🚀 Major Platform Integrations

• 🎭 Replicate: Access to open-source models including Llama 2, CodeLlama, Mistral, and more

• 🌟 AI21 Labs: Jamba 1.5 models with advanced reasoning capabilities

• 🎨 Stability AI: StableLM models including code-specialized variants

• 🔥 Fireworks AI: Ultra-fast inference for popular open-source models

• 🚀 Anyscale: Ray-powered LLM serving with enterprise-grade reliability

🆕 Enhanced Existing Platform Support

• 🤖 Mistral AI: Direct access to Mistral's latest models including mistral-large-latest and codestral-latest

• 🔗 Together AI: Access to 200+ open-source models with fast inference

• 🧠 Cohere: Enterprise-grade language models with Command R+ and Command R

• ⚡ Groq Fast: Ultra-fast inference API for lightning-quick responses

• 🔍 Perplexity AI: Web-connected AI with real-time search capabilities

🔧 Previous Bug Fixes (v3.0)

• Fixed HuggingFace Models: Completely rebuilt HuggingFace integration with advanced retry logic, better model format detection, and comprehensive error handling

• Fixed Gemini Blank Responses: Enhanced Gemini conversation handling to prevent empty responses in long chats with smart fallback and retry mechanisms

• Improved Error Handling: Better error messages with helpful suggestions for troubleshooting

🤖 HuggingFace Integration (Enhanced)

Access any of the 800,000+ models on HuggingFace Hub via their Inference API with improved reliability:

• meta-llama/Llama-3.1-8B-Instruct - Fast and reliable

• meta-llama/Llama-3.1-70B-Instruct - Powerful reasoning

• mistralai/Mistral-7B-Instruct-v0.3 - Efficient French-developed model

• Qwen/Qwen2.5-7B-Instruct - Alibaba's latest model

🧠 DeepSeek Models

Get opinions from DeepSeek's powerful reasoning models:

• deepseek-chat (DeepSeek-V3) - Fast and efficient

• deepseek-reasoner (DeepSeek-R1) - Advanced reasoning

🤔 Grok 4 Thinking

Access xAI's thinking models with enhanced reasoning:

• grok-4 - Latest flagship model

• grok-3-thinking - Step-by-step reasoning model, last gen

• grok-3-mini - Lightweight thinking model with reasoning_effort control, last ge

🎭 Group Discussions (Enhanced)

Start multi-AI discussions where models can see and respond to each other's input:

> "Start a group discussion about the future of AI with GPT-4.1, Claude-4, Mistral, and Perplexity"

🔧 Enhanced Performance

• Longer responses: Increased max_tokens (4000 default) to prevent cut-off responses

• Better error handling: More robust API interactions with exponential backoff

• Conversation persistence: Enhanced memory management with better context handling

• Smart retry logic: Automatic retries with progressive delays for better reliability

📋 Requirements

• Python 3.8+ - Programming language runtime

• Claude Desktop or Claude Code - Choose your preferred MCP integration

• API Keys - For any combination of the 16 supported AI platforms

📦 Installation Guide

🚀 Quick Start

1. 📥 Clone the repository

   git clone https://github.com/ProCreations-Official/second-opinion.git
   cd second-opinion

2. ⚙️ Install dependencies

   pip install -r requirements.txt

3. 🔑 Get API Keys

   Platform	Link	Required
   OpenAI	platform.openai.com	⭐ Popular
   Gemini	aistudio.google.com	⭐ Popular
   Grok	x.ai	🔥 Best benchmarks
   Claude	anthropic.com	🧠 Advanced
   HuggingFace	huggingface.co	🤗 800k+ Models
   DeepSeek	deepseek.com	🔬 Reasoning
   OpenRouter	openrouter.ai	🌍 200+ Models via One API
   Mistral	console.mistral.ai	🇫🇷 European, fast
   Together AI	api.together.xyz	🔗 200+ Models
   Cohere	dashboard.cohere.com	🏢 Enterprise
   Groq	console.groq.com	⚡ Ultra-Fast
   Perplexity	perplexity.ai	🔍 Web Search
   Replicate	replicate.com	🎭 Open Source
   AI21 Labs	studio.ai21.com	🧬 Jamba Models
   Stability AI	platform.stability.ai	🎨 StableLM
   Fireworks AI	fireworks.ai	🔥 Fast Inference
   Anyscale	console.anyscale.com	🚀 Ray Serving

4. 🔧 Choose Your Integration Method

   Select the method that matches your Claude setup:

For Claude Desktop Users

Add this to your Claude Desktop MCP configuration:

{
  "mcpServers": {
    "second-opinion": {
      "command": "python3",
      "args": ["/path/to/your/main.py"],
      "env": {
        "OPENAI_API_KEY": "your_openai_key_here",
        "GEMINI_API_KEY": "your_gemini_key_here",
        "GROK_API_KEY": "your_grok_key_here",
        "CLAUDE_API_KEY": "your_claude_key_here",
        "HUGGINGFACE_API_KEY": "your_huggingface_key_here",
        "DEEPSEEK_API_KEY": "your_deepseek_key_here",
        "OPENROUTER_API_KEY": "your_openrouter_key_here",
        "MISTRAL_API_KEY": "your_mistral_key_here",
        "TOGETHER_API_KEY": "your_together_key_here",
        "COHERE_API_KEY": "your_cohere_key_here",
        "GROQ_FAST_API_KEY": "your_groq_key_here",
        "PERPLEXITY_API_KEY": "your_perplexity_key_here",
        "REPLICATE_API_TOKEN": "your_replicate_key_here",
        "AI21_API_KEY": "your_ai21_key_here",
        "STABILITY_API_KEY": "your_stability_key_here",
        "FIREWORKS_API_KEY": "your_fireworks_key_here",
        "ANYSCALE_API_KEY": "your_anyscale_key_here"
      }
    }
  }
}

💡 Note: You only need to add API keys for the services you want to use. Missing keys will simply disable those specific features.

🔄 Restart Claude Desktop after configuration.

🎯 For Claude Code CLI Users

📦 Prerequisites

First, ensure Claude Code CLI is installed globally:

npm install -g @anthropic-ai/claude-code

🚀 Installation Methods

Use the claude mcp add command to add the Second Opinion server:

# Navigate to your second-opinion directory
cd /path/to/your/second-opinion

# Add the MCP server with environment variables (use -e for each API key)
claude mcp add second-opinion -s user \
  -e OPENAI_API_KEY=your_openai_key_here \
  -e GEMINI_API_KEY=your_gemini_key_here \
  -e GROK_API_KEY=your_grok_key_here \
  -e CLAUDE_API_KEY=your_claude_key_here \
  -e HUGGINGFACE_API_KEY=your_huggingface_key_here \
  -e DEEPSEEK_API_KEY=your_deepseek_key_here \
  -e OPENROUTER_API_KEY=your_openrouter_key_here \
  -e MISTRAL_API_KEY=your_mistral_key_here \
  -e TOGETHER_API_KEY=your_together_key_here \
  -e COHERE_API_KEY=your_cohere_key_here \
  -e GROQ_FAST_API_KEY=your_groq_key_here \
  -e PERPLEXITY_API_KEY=your_perplexity_key_here \
  -e REPLICATE_API_TOKEN=your_replicate_key_here \
  -e AI21_API_KEY=your_ai21_key_here \
  -e STABILITY_API_KEY=your_stability_key_here \
  -e FIREWORKS_API_KEY=your_fireworks_key_here \
  -e ANYSCALE_API_KEY=your_anyscale_key_here \
  -- /path/to/your/second-opinion/run.sh

💡 Quick Setup: You only need to include -e flags for the API keys you have. For example, if you only have OpenAI and Gemini keys:

claude mcp add second-opinion -s user \
  -e OPENAI_API_KEY=your_openai_key_here \
  -e GEMINI_API_KEY=your_gemini_key_here \
  -- /path/to/your/second-opinion/run.sh

Alternatively, you can manually add the server to your .claude.json file:

{
  "mcpServers": {
    "second-opinion": {
      "type": "stdio",
      "command": "/path/to/your/second-opinion/run.sh",
      "env": {
        "OPENAI_API_KEY": "your_openai_key_here",
        "GEMINI_API_KEY": "your_gemini_key_here",
        "GROK_API_KEY": "your_grok_key_here",
        "CLAUDE_API_KEY": "your_claude_key_here",
        "HUGGINGFACE_API_KEY": "your_huggingface_key_here",
        "DEEPSEEK_API_KEY": "your_deepseek_key_here",
        "OPENROUTER_API_KEY": "your_openrouter_key_here",
        "MISTRAL_API_KEY": "your_mistral_key_here",
        "TOGETHER_API_KEY": "your_together_key_here",
        "COHERE_API_KEY": "your_cohere_key_here",
        "GROQ_FAST_API_KEY": "your_groq_key_here",
        "PERPLEXITY_API_KEY": "your_perplexity_key_here",
        "REPLICATE_API_TOKEN": "your_replicate_key_here",
        "AI21_API_KEY": "your_ai21_key_here",
        "STABILITY_API_KEY": "your_stability_key_here",
        "FIREWORKS_API_KEY": "your_fireworks_key_here",
        "ANYSCALE_API_KEY": "your_anyscale_key_here"
      }
    }
  }
}

🔧 Why Use run.sh?

✅ Verification

Check that your MCP server is properly installed:

claude mcp list

You should see second-opinion in the list of available MCP servers.

🔑 Environment Variables: You only need to add API keys for the services you want to use. Missing keys will simply disable those specific AI platforms. The server will work with any combination of available API keys.

🚀 Premium AI Platforms

Featured Models:

🌟 Special: Access to any model on HuggingFace Hub that supports text generation

🔬 Specialized AI Platforms

🌟 Special: Access to 200+ models from multiple providers through a single OpenRouter API

Cohere (Enterprise-grade)

Groq (Ultra-fast inference)

Perplexity AI (Web-connected)

Replicate (Open-source hosting)

AI21 Labs (Advanced reasoning)

Stability AI (StableLM family)

Fireworks AI (Ultra-fast inference)

Anyscale (Ray-powered serving)

████████████████████████████████████████████████████████████████████████████████

💡 Usage Examples

🎭 NEW: Personality Examples

😤 "Give me an honest opinion about this code" (brutally frank feedback)

💕 "I need some encouragement with this project" (supportive girlfriend mode)

🏆 "Help me stay motivated to finish this task" (motivational coach)

🧙 "What's the deeper meaning behind this design pattern?" (ancient wisdom)

🎨 "Think of a creative solution to this problem" (innovative thinking)

🤖 "Just give me the best available opinion" (automatic smart selection)

🔥 Popular Requests

💬 "Get a second opinion from GPT-4.1 on this coding approach"

🤔 "What would Grok-4 think about this solution?" (NEW: Latest model)

⚖️ "Compare how Claude-4-opus and gemini-2.5-flash would solve this problem"

🤗 "Get an opinion from meta-llama/Llama-3.1-70B-Instruct on HuggingFace"

🌍 "Get an OpenRouter opinion from anthropic/claude-3-5-sonnet-20241022"

🧠 "What does DeepSeek-reasoner think about this math problem?"

🇫🇷 "Ask Mistral-large-latest to review my code architecture"

⚡ "Get a fast response from Groq's llama-3.1-8b-instant model"

🌐 "Use Perplexity's web search to research the latest AI developments"

🏢 "What does Cohere's command-r-plus think about this business strategy?"

🔗 "Get Together AI's Llama-405B opinion on this complex problem"

🎭 Advanced Features

🗣️ "Start a group discussion about AI ethics with GPT-4.1, Claude-4, Mistral, and Perplexity"

📊 "Cross-platform comparison of this algorithm across all 16 available platforms"

🎭 "Get a Replicate opinion from meta/llama-2-70b-chat on this open-source approach"

🧬 "What does AI21's Jamba-1.5-large think about this reasoning problem?"

🎨 "Ask Stability AI's StableLM about this code optimization"

🔥 "Get a super-fast response from Fireworks AI's Llama model"

🚀 "Use Anyscale's enterprise-grade Llama serving for this complex task"

🔧 Available Tools

Single Model Opinions

• get_openai_opinion - Get opinion from any OpenAI model

• get_gemini_opinion - Get opinion from any Gemini model (enhanced with better conversation handling)

• get_grok_opinion - Get opinion from any Grok model (includes thinking models)

• get_claude_opinion - Get opinion from any Claude model

• get_huggingface_opinion - Get opinion from any HuggingFace model (enhanced with better reliability)

• get_deepseek_opinion - Get opinion from DeepSeek models

• get_openrouter_opinion - Get opinion from 200+ models via OpenRouter (NEW)

• get_mistral_opinion - Get opinion from Mistral AI models (NEW)

• get_together_opinion - Get opinion from Together AI's 200+ models (NEW)

• get_cohere_opinion - Get opinion from Cohere enterprise models (NEW)

• get_groq_fast_opinion - Get ultra-fast responses from Groq (NEW)

• get_perplexity_opinion - Get web-connected AI responses

• get_replicate_opinion - Get opinion from Replicate's open-source models (NEW)

• get_ai21_opinion - Get opinion from AI21 Labs' Jamba models (NEW)

• get_stability_opinion - Get opinion from Stability AI's StableLM models (NEW)

• get_fireworks_opinion - Get ultra-fast responses from Fireworks AI (NEW)

• get_anyscale_opinion - Get enterprise-grade responses from Anyscale (NEW)

Model Comparisons

• compare_openai_models - Compare multiple OpenAI models

• compare_gemini_models - Compare multiple Gemini models

• compare_grok_models - Compare multiple Grok models

• compare_claude_models - Compare multiple Claude models

🎭 NEW: Personality & Smart Default Tools

• get_personality_opinion - Get AI responses with specific personality (honest, gf, coach, wise, creative)

• get_default_opinion - Automatically uses the best available model (Grok-4 → Gemini Pro → GPT-4.1)

• list_personalities - See all available AI personalities and their descriptions

Cross-Platform Features

• cross_platform_comparison - Compare across all 17 AI platforms: OpenAI, Gemini, Grok, Claude, HuggingFace, DeepSeek, OpenRouter, Mistral, Together AI, Cohere, Groq Fast, Perplexity, Replicate, AI21 Labs, Stability AI, Fireworks AI & Anyscale

• group_discussion - Multi-round discussions between AI models with shared context (supports all platforms)

Conversation Management

• list_conversation_histories - See active conversation threads

• clear_conversation_history - Reset conversation memory for specific models

🧠 Advanced Features

GPT 5 Thinking Models

For deeper reasoning, use thinking models:

> "Get a GPT 5 thinking opinion on this complex math problem with high reasoning effort"

This will use GPT 5 with reasoning effort set to high.

Group Discussions

Create AI debates and collaborative problem-solving:

> "Start a group discussion about renewable energy solutions with 3 rounds between GPT-4.1, Claude-4, Gemini, and DeepSeek"

Each AI can see previous responses and build on the discussion.

HuggingFace Model Access

Access cutting-edge open source models:

> "Get an opinion from Qwen/Qwen3-30B-A3B-Instruct-2507 about chatbot design patterns"

may be broken Perfect for testing specialized models or comparing open source vs proprietary AI.

🔒 Security

Your API keys stay private on your machine. The MCP server only sends model responses to the client, never your credentials.

🛟 Troubleshooting

Import errors: Ensure you've installed all dependencies with pip install -r requirements.txt

API errors: Check that your API keys are correct and active

Server not connecting: Verify the file path in your MCP configuration

Cut-off responses: The new version uses 4000 max_tokens by default to prevent truncation

HuggingFace timeouts: Some models may take time to load. Try again after a few moments.

Model not available: Check if the HuggingFace model supports text generation or chat completion

🚀 What's New in v5.0

• ✅ 🎭 AI Personality System: 5 distinct personalities (honest, gf, coach, wise, creative) with optimized model selection

• ✅ 🧠 Intelligent Model Selection: Quality-ranked models with Grok-4 as top priority, smart defaults

• ✅ 🏗️ Modular Architecture: Refactored from 51k+ token monolith into 5 maintainable modules

• ✅ 📊 JSON Configuration: Easy model priority updates via configuration files

• ✅ 🤖 Always-Available Tools: Personality and default tools work with any provider setup

• ✅ ⚡ Enhanced Performance: Optimized model selection and better error handling

• ✅ 🔄 Future-Proof: Easy to add new models and update priorities

Previous Updates (v4.0)

• ✅ 5 New Major AI Platforms: Replicate, AI21 Labs, Stability AI, Fireworks AI, Anyscale

• ✅ 16 Total Platforms: Now supporting 16 different AI platforms and 800,000+ models

• ✅ Advanced Reasoning: AI21 Labs' Jamba models with state-space architecture

• ✅ Ultra-Fast Inference: Fireworks AI for blazing-fast open model serving

Previous Improvements (v3.0)

• ✅ Major Bug Fixes: Fixed HuggingFace empty responses and Gemini blank chat issues

• ✅ Enhanced HuggingFace: Completely rebuilt with advanced retry logic and better error handling

• ✅ Improved Gemini: Smart conversation handling prevents blank responses in long chats

• ✅ Web-Connected AI: Perplexity AI with real-time search capabilities

• ✅ Enterprise Models: Cohere's command models for business use cases

🤝 Contributing

Issues and pull requests welcome! This is an open-source project for the AI community.

▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓

🧠✨ Built for developers who want access to the entire AI ecosystem at their fingertips

⭐ Star us on GitHub • 🍴 Fork the project • 💖 Contribute to the future of AI

▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓