https://github.com/mitsudoai/guru-pk-mcp
Multi-round AI expert debate system via MCP - Three thought leaders engage in deep discussions on any topic
https://github.com/mitsudoai/guru-pk-mcp
ai-agents ai-reasoning critical-thinking expert-debate guru-pk infographics mcp-server tufte-design
Last synced: 12 days ago
JSON representation
Multi-round AI expert debate system via MCP - Three thought leaders engage in deep discussions on any topic
- Host: GitHub
- URL: https://github.com/mitsudoai/guru-pk-mcp
- Owner: MitsudoAI
- License: mit
- Created: 2025-06-28T09:53:17.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2025-08-21T07:31:57.000Z (about 2 months ago)
- Last Synced: 2025-09-04T00:45:29.785Z (about 1 month ago)
- Topics: ai-agents, ai-reasoning, critical-thinking, expert-debate, guru-pk, infographics, mcp-server, tufte-design
- Language: HTML
- Homepage: https://mitsudoai.github.io/guru-pk-mcp/
- Size: 1.6 MB
- Stars: 5
- Watchers: 0
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# π Language Versions | ε€θ―θ¨ηζ¬ | θ¨θͺγγΌγΈγ§γ³
π **[English](README.md)** | π¨π³ **[δΈζ](README_zh.md)** | π―π΅ **[ζ₯ζ¬θͺ](README_ja.md)**
---
## Guru-PK MCP Intelligent Expert Debate System
An AI expert debate system based on local MCP (Model Context Protocol), featuring **dynamic expert generation architecture** that intelligently creates the most suitable expert combinations based on questions for multi-round intellectual confrontation.
## β¨ Core Features
- π **Dynamic Expert Generation** - Completely question-driven, generating dedicated expert combinations each time
- π **Unlimited Expert Pool** - Breaking fixed expert limitations, supporting expert generation in any domain
- π **Multi-Round PK Process** - Independent Thinking β Cross-Debate β Final Positions β Wisdom Synthesis
- π¨ **Tufte-Style Infographics** - Transform expert debates into single-page dynamic infographics strictly following data visualization master Edward Tufte's design principles
- π€ **Intelligent Division Architecture** - MCP Host-side LLM handles intelligent analysis, MCP Server-side provides process guidance## π Online Demo
**π [View Infographic Demo](https://mitsudoai.github.io/guru-pk-mcp/)**
This webpage displays Tufte-style dynamic infographics created using this MCP tool, intuitively showcasing the powerful capabilities of the expert debate system.
## π Quick Installation
### 1. Install Dependencies
**Method 1: Using Installation Script (Recommended)**
**macOS/Linux:**
```bash
curl -LsSf https://astral.sh/uv/install.sh | sh
```**Windows:**
```powershell
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
```**Method 2: Install with pip (All Platforms)**
```bash
pip install uv
```**Method 3: Download Installation Package**
Download the installation package for your platform from [UV Releases](https://github.com/astral-sh/uv/releases)
### 2. Configure MCP Client
**Recommended Method: Install from PyPI**
```json
{
"mcpServers": {
"guru-pk": {
"command": "uvx",
"args": ["--from", "guru-pk-mcp", "guru-pk-mcp-server"],
"env": {
"DATA_DIR": "~/.guru-pk-data" // macOS/Linux: ~ directory, Windows: %USERPROFILE% directory
}
}
}
}
```> **Update Instructions**:
>
> - When you need to update `guru-pk-mcp` to the latest version, run:
>
> ```bash
> uvx pip install --upgrade guru-pk-mcp
> ```
>
> - This command fetches and installs the latest released version from PyPI
> - If you encounter cache issues, you can force refresh:
>
> ```bash
> uvx --refresh-package guru-pk-mcp --from guru-pk-mcp python -c "print('β UVX cache refreshed')"
> ```
>
> **Notes**:
>
> - macOS users might need to use the full path: `/Users/{username}/.local/bin/uvx`
> - Windows users: `~` automatically resolves to user home directory (e.g., `C:\\Users\\{username}`), no manual modification needed**Development Method: Install from Source**
```json
{
"mcpServers": {
"guru-pk": {
"command": "uvx",
"args": ["--from", "/path/to/guru-pk-mcp", "guru-pk-mcp-server"],
"env": {
"DATA_DIR": "~/.guru-pk-data" // macOS/Linux: ~ directory, Windows: %USERPROFILE% directory
}
}
}
}
```> **Local Development Instructions**:
>
> - For local development scenarios, if you need to refresh uvx cache, use `make refresh-uvx`
> - This command forces UVX to reinstall the local package, ensuring the use of latest code changes## Getting Started
Restart your MCP client, enter `guru_pk_help` to get help, or directly ask questions to start expert debates!
```javascript
// 1. Natural language questions (most recommended usage)
Are there any directions in the field of generative AI that are particularly suitable for individual entrepreneurship? Please have three experts PK// 2. Intelligent candidate expert generation (system automatic execution)
start_pk_session: Are there any directions in the field of generative AI that are particularly suitable for individual entrepreneurship?// 3. Intelligent candidate expert generation (user limits expected expert scope)
start_pk_session: Are there any directions in the field of generative AI that are particularly suitable for individual entrepreneurship? Find two AI field experts and one famous individual entrepreneur to debate
```### π‘ Usage Tips
**Starting Debates**:
- π€ **`start_pk_session: direct question`** - Default high-efficiency batch processing mode (recommended)
- π **`start_stepwise_pk_session: direct question`** - Traditional step-by-step dialogue mode**Tool Functions**:
- π `guru_pk_help` - Get system introduction and detailed help
- π `export_session` - Export session as Markdown file
- π¨ `export_session_as_infographic` - Export session as Tufte-style single-page dynamic infographic
- π `export_enhanced_session` - Export enhanced analysis report
- π `set_language` - Set expert reply language### π± Compatibility
Supports all MCP-compatible applications: Claude Desktop, Cursor, TRAE, DeepChat, Cherry Studio, etc.
### π― Recommended Configuration
**Most Recommended MCP Hosts**:
- π° **Subscription-based MCP Hosts calculated by user requests** - Such as Cursor and overseas TRAE
- π **Advantages**:
- Significant cost advantages: subscription billing calculated by user requests, not API call counts or token billing
- Claude models have the best MCP support with excellent instruction-following capabilities### β οΈ Not Recommended Configuration
- π« **TRAE Domestic Version** - Built-in domestic models have sensitive word censorship issues that may interrupt expert debate processes, affecting user experience
## π οΈ Technical Architecture
**Intelligent Division Principles**:
- π§ **MCP Host-side LLM**: Responsible for complex semantic analysis and intelligent generation
- π§ **MCP Server-side**: Provides concise process control and data management### Dynamic Expert Generation Flow
```mermaid
flowchart TD
A[π€ Raise Question] --> B[π§ Intelligent Analysis]
B --> C[π₯ Generate Candidates]
C --> D[π Start Debate]
A1[Ask system directly about any topic]
B1[MCP Host-side LLM deeply analyzes question characteristics]
C1[Dynamically create 3 most relevant experts]
D1[Launch multi-round PK process]
A -.-> A1
B -.-> B1
C -.-> C1
D -.-> D1
style A fill:#e1f5fe
style B fill:#f3e5f5
style C fill:#e8f5e8
style D fill:#fff3e0
```### π Debate Flow
**Two Debate Modes**:
π **Batch Processing Mode** (`start_pk_session`) - **Default Recommended**
- β‘ High Efficiency: Generate all expert answers in one round, saving about 60% time
- π― Use Cases: Rapidly obtain multi-perspective analysis, efficient decision supportπ **Stepwise Mode** (`start_stepwise_pk_session`) - Traditional Experience
- π Interactive: Experts speak sequentially, allowing real-time adjustment and deep exploration
- π― Use Cases: Deep contemplation, enjoying the complete debate process**4-Round Debate Flow**:
```mermaid
flowchart TD
A[π€ Independent Thinking] --> B[βοΈ Cross-Debate]
B --> C[π― Final Positions]
C --> D[π§ Wisdom Synthesis]
A1[Each expert independently analyzes the problem]
B1[Experts mutually question and learn from each other]
C1[Form their respective refined solutions]
D1[Ultimate answer integrating all perspectives]
A -.-> A1
B -.-> B1
C -.-> C1
D -.-> D1
B --> B2[Multi-round Interaction]
B2 --> B
style A fill:#e3f2fd
style B fill:#fce4ec
style C fill:#e8f5e8
style D fill:#fff8e1
style B2 fill:#f3e5f5
```