Think MCP Host (AI·Zen·Love)

Think MCP Host (AI·Zen·Love) is a Model Context Protocol (MCP) based intelligent agent application that supports various types of large language models, including standard conversational models (LLM), vision language models (VLM), and reasoning models.

Features

• Complete MCP (Model Context Protocol) Implementation

  • Full MCP architecture support (Host/Client/Server)
  • Comprehensive MCP resource types support
    • Resources: Dynamic integration of external content
    • Prompts: Template-based system prompts
    • Tools: AI-powered function calls
  • Dynamic MCP command insertion anywhere in conversations
    • Seamless integration of resources into context
    • On-demand prompt template usage
    • Direct tool execution within chat
  • Standalone MCP tool execution support

• Extensive Model Support

  • LLM (Language Models)
    • Text conversations and content generation
    • Programming and code assistance
    • Document writing and analysis
  • VLM (Vision Language Models)
    • Image understanding and analysis
    • Visual content processing
  • Reasoning Models
    • Complex logical analysis
    • Professional domain reasoning
  • Multiple provider support (DeepSeek, OpenAI, OpenRouter, etc.)

• Advanced Conversation Management

  • Automatic conversation history saving
  • Manual save options with countdown timer
  • Historical conversation loading
  • Multiple export format support

• System Features

  • Rich Terminal Interface
    • Beautiful markdown rendering in terminal
    • Syntax highlighting for code blocks
    • Unicode and emoji support
    • Interactive command suggestions
  • Cross-Platform Support
    • Full functionality on Windows, macOS, and Ubuntu
    • Native installation support for each platform
    • Consistent user experience across systems
  • Command-line interface
  • Debug mode support
  • Flexible exit options with save/discard choices

Usage Guide

Running Mode Selection

The program supports two main running modes:

1. Chat Mode (Default)

   • Used for natural language dialogue
   • Supports multiple LLM models
   • Can use MCP enhancement features

2. Tool Mode

   • Used for using specific AI tools
   • Directly calls functions provided by MCP server

Detailed Usage Process

1. Select Running Mode

• After program startup, you will be prompted to select running mode
• Enter 1 to select Chat mode
• Enter 2 to select Tool mode

2. Chat Mode Setup Process

   1. Select LLM Model

      • System will display available model list
      • Enter corresponding number to select model
      • Supported models include DeepSeek, Silicon Flow, Volcano Engine, etc.

   2. Choose Start Method

      • Option 1: Set system prompt, then start new conversation
      • Option 2: Directly start new conversation (default)
      • Option 3: Load historical conversation

   3. System Prompt Setting (if Option 1 was selected)

      • Can input custom system prompt
      • Supports using ->mcp command to insert MCP resources

   4. Load Historical Conversation (if Option 3 was selected)

      • System will display available historical conversation list
      • Select conversation record to load

3. Tool Mode Setup Process

   1. Select MCP Client

      • System will display available MCP client list
      • Select client to use

   2. Select Tool

      • Displays tool list provided by selected client
      • Select specific tool to use

   3. Execute Tool

      • Provide necessary parameters according to tool requirements
      • View tool execution results

   4. Continue or Exit

      • Choose whether to continue using other tools
      • Can switch back to Chat mode at any time

Basic Chat Mode

1. Start Conversation
   • Directly input text to converse
   • Use Ctrl+C to exit program

MCP Enhanced Mode

During conversation, you can use the ->mcp command to use MCP's enhancement features. Steps are as follows:

1. Activate MCP Command

   • Input ->mcp alone and press Enter in conversation
   • System will guide you through subsequent selections

2. Select MCP Client

   • System will display available MCP client list
   • Select client to use

3. Select MCP Feature Type System will prompt you to select one of three types:

   1. Resources

      • Input 1 to select
      • Used for selecting and referencing external resources (like images, documents, etc.)
      • Returns format: ->mcp_resources[client_name]:resourceURI

   2. Prompts

      • Input 2 to select
      • Used for selecting predefined prompt templates
      • Returns format: ->mcp_prompts[client_name]:prompt_name{parameters}

   3. Tools

      • Input 3 to select
      • Used for selecting and using specific AI tools
      • Returns format: ->mcp_tools[client_name]:tool_name{parameters}

4. Complete Selection

   • After selection is complete, system will insert corresponding MCP command in conversation
   • You can continue editing message, or send directly

5. Select Running Mode

• After program startup, you will be prompted to select running mode
• Enter 1 to select Chat mode
• Enter 2 to select Tool mode

2. Chat Mode Setup Process

   1. Select LLM Model

      • System will display available model list
      • Enter corresponding number to select model
      • Supported models include DeepSeek, Silicon Flow, Volcano Engine, etc.

   2. Choose Start Method

      • Option 1: Set system prompt, then start new conversation
      • Option 2: Directly start new conversation (default)
      • Option 3: Load historical conversation

   3. System Prompt Setting (if Option 1 was selected)

      • Can input custom system prompt
      • Supports using ->mcp command to insert MCP resources

   4. Load Historical Conversation (if Option 3 was selected)

      • System will display available historical conversation list
      • Select conversation record to load

3. Tool Mode Setup Process

   1. Select MCP Client

      • System will display available MCP client list
      • Select client to use

   2. Select Tool

      • Displays tool list provided by selected client
      • Select specific tool to use

   3. Execute Tool

      • Provide necessary parameters according to tool requirements
      • View tool execution results

   4. Continue or Exit

      • Choose whether to continue using other tools
      • Can switch back to Chat mode at any time

Basic Chat Mode

1. Start Conversation
   • Directly input text to converse
   • Use Ctrl+C to exit program

MCP Enhanced Mode

During conversation, you can use the ->mcp command to use MCP's enhancement features. Steps are as follows:

1. Activate MCP Command

   • Input ->mcp alone and press Enter in conversation
   • System will guide you through subsequent selections

2. Select MCP Client

   • System will display available MCP client list
   • Select client to use

3. Select MCP Feature Type System will prompt you to select one of three types:

   1. Resources

      • Input 1 to select
      • Used for selecting and referencing external resources (like images, documents, etc.)
      • Returns format: ->mcp_resources[client_name]:resourceURI

   2. Prompts

      • Input 2 to select
      • Used for selecting predefined prompt templates
      • Returns format: ->mcp_prompts[client_name]:prompt_name{parameters}

   3. Tools

      • Input 3 to select
      • Used for selecting and using specific AI tools
      • Returns format: ->mcp_tools[client_name]:tool_name{parameters}

4. Complete Selection

   • After selection is complete, system will insert corresponding MCP command in conversation
   • You can continue editing message, or send directly

Installation and Running

Development Installation

Before installing from package repositories, you can install the project directly from source for development:

# Clone the repository
git clone https://github.com/thinkthinking/think-mcp-host.git
cd think-mcp-host

# Create and activate virtual environment
python -m venv .venv
source .venv/bin/activate  # On Linux/macOS
# or
.venv\Scripts\Activate.ps1  # On Windows with PowerShell

# Install in development mode with pip
pip install -e .
# or with uv (recommended)
uv pip install -e .

Windows

1. Installation methods
   • Download and double-click AI-Zen-Love.exe
   • Or install and run via command line:

# Install uv using pip
python -m pip install uv

# Clone the project and install
git clone https://github.com/thinkthinking/think-mcp-host.git
cd think-mcp-host
python -m venv .venv
.venv\Scripts\Activate.ps1
uv pip install -e .

2. Configuration file locations
   • LLM configuration: C:\Users\your-username\.think-llm-client\config\servers_config.json
   • MCP configuration: C:\Users\your-username\.think-mcp-client\config\mcp_config.json
   • History records: C:\Users\your-username\.think-mcp-host\command_history\

macOS

1. Installation methods
   • Download and double-click AI-Zen-Love.app
   • Or install and run via terminal:

# Install uv
python3 -m pip install uv

# Clone the project and install
git clone https://github.com/thinkthinking/think-mcp-host.git
cd think-mcp-host
python3 -m venv .venv
source .venv/bin/activate
uv pip install -e .

2. Configuration file locations
   • LLM configuration: /Users/your-username/.think-llm-client/config/servers_config.json
   • MCP configuration: /Users/your-username/.think-mcp-client/config/mcp_config.json
   • History records: /Users/your-username/.think-mcp-host/command_history/

Configuration Details

Model Configuration

The project supports three types of models:

1. LLM (Language Models)

   • Used for: Text conversations, code writing, document generation
   • Examples: DeepSeek Chat, GPT-4

2. VLM (Vision Language Models)

   • Used for: Image understanding and analysis
   • Examples: GPT-4-Vision, Qwen-VL-Plus

3. Reasoning Models

   • Used for: Complex reasoning and professional analysis
   • Examples: DeepSeek Reasoner, DeepSeek-R1

LLM Configuration

The configuration file uses JSON format and needs to be configured according to different model types:

{
  "llm": {
    "providers": {
      "deepseek": {
        "api_key": "<DEEPSEEK_API_KEY>",
        "api_url": "https://api.deepseek.com",
        "model": {
          "deepseek-chat": {
            "max_completion_tokens": 8192
          }
        }
      }
    }
  },
  "vlm": {
    "providers": {
      "openai": {
        "api_key": "<OPENAI_API_KEY>",
        "api_url": "https://api.openai.com/v1",
        "model": {
          "gpt-4-vision": {
            "max_completion_tokens": 4096
          }
        }
      }
    }
  },
  "reasoning": {
    "providers": {
      "deepseek": {
        "api_key": "<DEEPSEEK_API_KEY>",
        "api_url": "https://api.deepseek.com",
        "model": {
          "deepseek-reasoner": {
            "max_completion_tokens": 8192,
            "temperature": 0.6
          }
        }
      }
    }
  }
}

Configuration explanation:

1. Choose the configuration area according to model type (llm/vlm/reasoning)
2. Multiple providers can be configured under each type Configuration instructions for different providers are as follows:

• DeepSeek Documentation: https://api-docs.deepseek.com/en/
• Silicon Flow Documentation: https://docs.siliconflow.cn/en/userguide/quickstart#4-siliconcloud-api-genai
• Volcano Engine Documentation: https://www.volcengine.com/docs/82379/1399008

3. Each provider needs to configure:
   • api_key: API key
   • api_url: API server address
   • model: Specific model configuration
     • max_completion_tokens: Maximum output length
     • temperature: Temperature parameter (optional)

MCP Server Configuration

MCP (Model Context Protocol) server configuration example:

{
  "mcpServers": {
    "think-mcp": {
      "command": "/opt/homebrew/bin/uv",
      "args": [
        "--directory",
        "/Users/thinkthinking/src_code/nas/think-mcp",
        "run",
        "think-mcp"
      ]
    }
  }
}

MCP Commands

The following MCP command formats can be used in conversations:

1. Interactive Selection

 ->mcp 

This will start an interactive selection interface, guiding you to choose:

• MCP client
• Operation type (Resources/Prompts/Tools)
• Specific resource/prompt/tool
• Related parameters (if needed)

2. Direct Usage

# Use resources
->mcp_resources[client_name]:resource_uri

# Use prompts
->mcp_prompts[client_name]:prompt_name{param1:value1,param2:value2}

# Use tools
->mcp_tools[client_name]:tool_name{param1:value1,param2:value2}

Examples:

# Use prompts
->mcp_prompts[think-mcp]:agent-introduction{agent_name:AI Assistant,agent_description:A friendly AI assistant}

# Use tools
->mcp_tools[think-mcp]:analyze_content{text:This is a test text}

Features

• Support for multiple MCP commands in the same input
• Commands can be edited and modified at any time
• Parameters support flexible key-value pair format
• Friendly error prompts

### MCP Commands

The following MCP command formats can be used in conversations:

1. Interactive Selection

```bash
 ->mcp 

This will start an interactive selection interface, guiding you to choose:

• MCP client
• Operation type (Resources/Prompts/Tools)
• Specific resource/prompt/tool
• Related parameters (if needed)

2. Direct Usage

# Use resources
->mcp_resources[client_name]:resource_uri

# Use prompts
->mcp_prompts[client_name]:prompt_name{param1:value1,param2:value2}

# Use tools
->mcp_tools[client_name]:tool_name{param1:value1,param2:value2}

Examples:

# Use prompts
->mcp_prompts[think-mcp]:agent-introduction{agent_name:AI Assistant,agent_description:A friendly AI assistant}

# Use tools
->mcp_tools[think-mcp]:analyze_content{text:This is a test text}

Features

• Support for multiple MCP commands in the same input
• Commands can be edited and modified at any time
• Parameters support flexible key-value pair format
• Friendly error prompts

Releasing New Versions

To release a new version, follow these steps:

1. Update the version number:

   • Update the version field in pyproject.toml
   • Follow Semantic Versioning

2. Commit changes:

git add pyproject.toml
git commit -m "chore: bump version to x.x.x"

3. Create version tag:

git tag vx.x.x
git push origin vx.x.x