上下文一致性管理服务

{
  "mcpServers": {
    "cc-mcp": {
      "autoApprove": [
        "process_user_message",
        "start_session",
        "get_debug_info", 
        "list_sessions",
        "get_session_stats",
        "export_context",
        "import_context"
      ],
      "disabled": false,
      "timeout": 60,
      "type": "stdio",
      "command": "uv",
      "args": [
        "run",
        "--directory",
        "/path/to/your/cc-mcp",
        "main.py"
      ],
      "env": {
        "CLASSIFIER_API_URL": "https://api.openai.com/v1/chat/completions",
        "CLASSIFIER_API_KEY": "your_openai_api_key_here",
        "CLASSIFIER_MODEL": "gpt-4o-mini"
      }
    }
  }
}

"env": {
  "CLASSIFIER_API_URL": "http://localhost:11434/v1/chat/completions",
  "CLASSIFIER_MODEL": "llama3.2"
}

Core Context (Long-term)    ──► Problem definitions & core objectives
    ↓
Evolving Context (Mid-term) ──► Constraints, decisions & refinements  
    ↓
Turn Context (Short-term)   ──► Recent conversation flow

User: "I want to build an AI assistant app"
AI: "Sure! Here are some general approaches..."

[10 messages later]
User: "Remember, budget is 500K yen, 3 months timeline"
AI: "What project are we talking about?" ❌

User: "I want to build an AI assistant app"
AI: "Great! Let me help you design this system..."

[10 messages later] 
User: "What about deployment options?"
AI: "For your AI assistant app (budget: 500K yen, 3-month timeline, security priority), here are deployment strategies that fit your constraints..." ✅

graph TD
    subgraph CC-MCP Server
        direction LR
        C[Intent Classifier]
        D[Context Store]
        E[Keyword Extractor]
    end

    A[MCP Client Application] -- MCP Tools --> B((main.py))
    B -- Manages --> C
    B -- Manages --> D
    B -- Manages --> E
    C -- Calls --> G[External LLM API]

# Clone the repository
git clone https://github.com/Beginnersguide138/cc-mcp.git
cd cc-mcp

# Install dependencies
uv sync

# Configure environment
cp .env.example .env
# Edit .env with your API credentials

# Fetch the latest changes from the remote repository
git fetch origin

# Pull the latest changes from the main branch
git pull origin main

# Update dependencies (if they have changed)
uv sync

# .env file example - Only 3 variables needed!
CLASSIFIER_API_URL=https://api.openai.com/v1/chat/completions
CLASSIFIER_API_KEY=sk-your_openai_api_key_here
CLASSIFIER_MODEL=gpt-3.5-turbo

# Alternative providers (see .env.example for full list):
# Azure OpenAI, Anthropic Claude, Google Gemini, Ollama supported

# For Ollama (Local LLM):
# CLASSIFIER_API_URL=http://localhost:11434/v1/chat/completions
# CLASSIFIER_API_KEY=not_required  # Ollama doesn't need auth; leave empty with the latest version
# CLASSIFIER_MODEL=llama3

# Start CC-MCP server with stdio transport (default for clients like Cline)
uv run main.py

# Install Ollama from https://ollama.com/
# Download and run a model
ollama run llama3

CLASSIFIER_API_URL=http://localhost:11434/v1/chat/completions
CLASSIFIER_API_KEY=not_required  # Ollama doesn't need auth; leave empty
CLASSIFIER_MODEL=llama3  # Or: mistral, gemma, phi, qwen, etc.

"env": {
  "CLASSIFIER_API_URL": "http://localhost:11434/v1/chat/completions",
  "CLASSIFIER_API_KEY": "not_required",
  "CLASSIFIER_MODEL": "llama3"
}

import asyncio
from cc_mcp import process_user_message

async def demo():
    # Define core problem
    result = await process_user_message(
        "I want to build an AI assistant for customer support"
    )
    print("Intent:", result["metadata"]["intent_classification"]["intent"])
    # Expected: ["PROBLEM_DEFINITION"]
    
    # Add constraints
    result = await process_user_message(
        "Budget is $50,000 and must be completed in 2 months"
    )
    print("Intent:", result["metadata"]["intent_classification"]["intent"])
    # Expected: ["CONSTRAINT_ADDITION"]
    
    # Ask unrelated question - context still preserved!
    result = await process_user_message(
        "What's the weather like today?"
    )
    # AI response will still maintain awareness of the support system project!

asyncio.run(demo())

# Export current conversation state
context_data = await export_context(session_id="my-session")
print(f"Core Problem: {context_data['core_problem']}")
print(f"Active Constraints: {len(context_data['constraints'])}")

# Clear context for fresh start
await clear_context(session_id="my-session")

# Import previously saved context
await import_context(json_state=context_data, session_id="my-session")

# Start new session
session_info = await start_session()
session_id = session_info["session_id"]

# Process messages in specific session
await process_user_message(
    message="Design a mobile app",
    session_id=session_id
)

# Check session statistics
stats = await get_session_stats(session_id=session_id)
print(f"Messages: {stats['total_messages']}")
print(f"Active Constraints: {stats['active_constraints']}")

# Extend intent classification in intent_classifier.py
CUSTOM_INTENTS = {
    "TECHNICAL_SPEC": "User is providing technical specifications",
    "BUDGET_CHANGE": "User is modifying budget constraints",
    "TIMELINE_UPDATE": "User is adjusting project timeline"
}

# Configure in context_store.py
CONTEXT_CONFIG = {
    "max_core_contexts": 3,      # Maximum core problems per session
    "max_evolving_items": 10,    # Maximum constraints/refinements
    "turn_history_limit": 5,     # Recent conversation turns to keep
    "auto_cleanup": True         # Automatic memory optimization
}

# Optimize for your use case
CLASSIFIER_PARAMS = {
    "temperature": 0.1,          # Lower for consistency
    "max_tokens": 150,          # Sufficient for classification
    "top_p": 0.9               # Focus on high-probability responses
}

MAIN_LLM_PARAMS = {
    "temperature": 0.7,          # Higher for creativity
    "max_tokens": 2000,         # Generous for detailed responses
    "presence_penalty": 0.1     # Encourage diverse vocabulary
}

# Clone your fork
git clone https://github.com/YOUR-USERNAME/cc-mcp.git
cd cc-mcp

# Install development dependencies
uv sync --group dev

# Run pre-commit hooks
pre-commit install

# Run the full test suite
uv run pytest