DeepMyst MCP Server

Intelligent LLM Optimization & Routing for Claude Desktop

Overview

DeepMyst MCP Server creates a seamless bridge between DeepMyst and Claude Desktop or any client through the Model Context Protocol (MCP). This integration allows Claude and other clients to harness DeepMyst's powerful optimization and routing capabilities while maintaining your familiar workflows.

Key Features

• Token Optimization - Reduce token usage by up to 75% while preserving response quality, directly lowering your API costs
• Smart Model Routing - Automatically select the optimal LLM for each specific query based on task requirements
• Combined Capabilities - Use both optimization and routing together for maximum efficiency

How It Works

• Token Optimization: DeepMyst identifies redundancies in prompts, intelligently compresses content, preserves key information, and maintains contextual meaning—all while significantly reducing token usage without sacrificing quality.

• Smart Routing: The system analyzes each query's category, complexity level, and required capabilities. It evaluates available models based on performance benchmarks, token cost, response latency, and capability support through a weighted scoring system.

Installation

Prerequisites

• Python 3.8 or higher
• UV - Fast Python package installer
• Claude Desktop - Latest version (or another MCP client)
• DeepMyst API key (get one from platform.deepmyst.com)

Installation Steps

1. Clone or download the DeepMyst MCP Server code:

# Create a directory for the server
mkdir DeepMyst-MCP
cd DeepMyst-MCP

# Download the server code
# (Or copy the code from the provided deepmyst_mcp.py file)

2. Install dependencies with UV:

# Create and activate a virtual environment (optional but recommended)
uv venv
source .venv/bin/activate  # On macOS/Linux
# OR
.venv\Scripts\activate     # On Windows

# Install dependencies
uv pip install mcp openai aiohttp

Configuration

Setting Up Your DeepMyst API Key

Environment Variable (Recommended):

export DEEPMYST_API_KEY="your-deepmyst-api-key"  # macOS/Linux
# OR
set DEEPMYST_API_KEY="your-deepmyst-api-key"     # Windows

Configuring Claude Desktop:

1. Open your Claude Desktop configuration file:

   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json

2. Add the DeepMyst MCP server configuration:

{
  "mcpServers": {
    "deepmyst": {
      "command": "uv",
      "args": [
        "run",
        "C:/Users/username/Documents/DeepMyst-MCP/deepmyst_mcp.py"
      ],
      "env": {
        "DEEPMYST_API_KEY": "your-deepmyst-api-key"
      }
    }
  }
}

Make sure to replace the file path and API key with your actual values, then save and restart Claude Desktop.

Tools and Capabilities

The DeepMyst MCP server provides several powerful tools:

1. Get Best Model for Query
   Analyzes your query and recommends the optimal model.

   Prompt: What's the best model for writing a poem about quantum physics?
   

2. Optimized Completion
   Generates a response with token optimization to reduce costs.

   Prompt: Analyze the attached book, using the optimized_completion tool.
   

3. Auto-Routed Completion
   Generates a response using the DeepMyst router to select the best model.

   Prompt: Generate a Python function to analyze sales data using auto_routed_completion.
   

4. Smart Completion
   Combines routing and optimization - first determines the best model for your query, then optionally applies token optimization.

   Prompt: Use smart_completion to explain how blockchain works in simple terms.
   

5. DeepMyst Completion
   The most flexible tool with all options configurable.

   Prompt: Use deepmyst_completion with GPT-4o, token optimization, and no routing to summarize this article.
   

Advanced Use Cases

Collaborative Multi-LLM Problem Solving

DeepMyst enables complex problems to be broken down and distributed across multiple specialized LLMs:

• Parallel Processing: Divide complex tasks into subtasks that different LLMs can solve simultaneously
• Expert Ensembles: Assign different aspects of a problem to models with specific strengths (e.g., GPT-4o for creative writing, Claude for logical reasoning, Gemini for mathematical analysis)
• Consensus Building: Route the same question to multiple models and synthesize their responses for more reliable answers
• Chain-of-Thought Enhancement: Use one model to generate a reasoning path, then have another model verify or improve it

Example: For a complex business strategy analysis, DeepMyst could route market research to data-focused models, creative ideation to generative specialists, and risk assessment to models with better reasoning capabilities.

Handling Large Context Windows

Efficiently manage and process large documents or complex conversations:

• Context Chunking: Break large documents into manageable segments, route each to the appropriate model, then synthesize the results
• Progressive Summarization: Use models with smaller context windows to summarize sections, then feed those summaries to models with broader context capabilities
• Priority Filtering: Intelligently identify and preserve the most relevant context while compressing less important information
• Context Management: Maintain conversation history effectively without exceeding token limits by dynamically compressing older turns

Example: When analyzing a 500-page legal document, DeepMyst can chunk the document, route sections to specialized legal analysis models, and progressively build a comprehensive analysis without hitting context limits.

Intelligent Task Routing

Match tasks to the most suitable models based on their specific capabilities:

• Capability Matching: Automatically identify task requirements and match them with models that excel in those areas
• Cost Optimization: Route simple queries to efficient, lower-cost models while reserving premium models for complex tasks
• Latency Management: Select models based on response time requirements for time-sensitive applications
• Specialization Routing: Direct domain-specific questions to models with the best performance in those fields

Example: A financial analysis workflow could route data processing to fast, efficient models, numerical analysis to math-specialized models, and final report generation to models with better writing capabilities.

Adaptive Learning and Improvement

Continuously improve routing decisions based on performance:

• Performance Tracking: Monitor response quality, token usage, and latency across different models for various task types
• Adaptive Routing: Refine routing decisions based on historical performance data
• A/B Testing: Automatically test different models on similar tasks to identify performance patterns
• Preference Learning: Adjust routing based on user feedback and preferences

Example: After observing that a specific model consistently performs better for creative writing tasks, DeepMyst can automatically adjust its routing to prefer that model for future creative assignments.

Supported LLM Providers

DeepMyst supports multiple LLM providers including:

• OpenAI: GPT-4o, GPT-4o-mini, etc.
• Anthropic: Claude 3.7 Sonnet, Claude 3.5 Sonnet, etc.
• Google: Gemini 2.0 Flash, Gemini 1.5 Pro, etc.
• Groq: Llama 3.1, Mixtral, etc.

Custom Configurations

You can customize tool behavior by passing specific parameters:

Prompt: Use deepmyst_completion with these parameters:
- base_model: gpt-4o-mini
- system_message: You are a helpful assistant specializing in finance
- optimize: true
- auto_route: false
- temperature: 0.3
- max_tokens: 2000

Resources

• Sign Up: platform.deepmyst.com - Create a DeepMyst account and get your API key
• Try DeepMyst: ask.deepmyst.com - Interactive demo to experience DeepMyst's capabilities
• Documentation: docs.deepmyst.com - Comprehensive guides and API documentation

Troubleshooting

Common Issues

Claude doesn't show the hammer icon:

• Make sure Claude Desktop is up to date
• Check your configuration file for syntax errors
• Restart Claude Desktop

API Key errors:

• Verify your DeepMyst API key is correct
• Ensure the key is properly set as an environment variable or in the config

Server connection issues:

• Check that all dependencies are installed
• Verify the path to deepmyst_mcp.py is correct
• Look for error messages in the terminal running the server

Logs

Check the DeepMyst MCP server logs for more detailed troubleshooting information. Logs are written to deepmyst.log in the same directory as the server script.

---

DeepMyst MCP Server is licensed under the MIT License