guru-pk

Guru-PK MCP Intelligent Expert Debate System dynamically generates expert candidates for each question and guides multi-round debates to synthesize a final, well-rounded answer. The host-side MCP analyzes the prompt, then the server-side component creates the most relevant experts and runs a four-phase debate: independent thinking, cross-debate, final positions, and wisdom synthesis. Users can start debates in batch mode for speed or opt into stepwise mode for deeper exploration. Useful tools include: guru_pk_help to read system information and options, export_session to save a debate as Markdown, export_session_as_infographic to produce a Tufte-style infographic, export_enhanced_session for a deeper analysis report, and set_language to tailor responses to a preferred language. Start a session with the built-in commands or natural language prompts like “start_pk_session: ...” to trigger the automatic expert generation and multi-round process.

Prerequisites:

• A working MCP client compatible with UV (Python/uv) installed on your system
• Network access to install or fetch the guru-pk-mcp-server package

Recommended installation (uvx-based approach):

# 1) Install UV if you don't have it yet (macOS/Linux)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows users (PowerShell):
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Alternatively, install via pip (All Platforms):

pip install uv

2. Install and configure the guru-pk MCP server (via UVX cache or local path):

# Recommended: install from PyPI using UVX (example from README)
uvx --from guru-pk-mcp guru-pk-mcp-server

3. If you are developing locally, you can point UVX to a local path:

uvx --from /path/to/guru-pk-mcp guru-pk-mcp-server

4. Ensure environment variable DATA_DIR is set if required by your setup, e.g.,

export DATA_DIR="~/.guru-pk-data"

5. Start the MCP server using the configuration shown in mcp_config:

# Example command, typically managed by your MCP client according to the config
uvx --from guru-pk-mcp guru-pk-mcp-server

Notes and tips:

• macOS/Linux users should place data in the home directory (as ~) or a user-specific data directory; Windows users will resolve %USERPROFILE% automatically where applicable.
• If you need to refresh the UVX cache for local development, use commands like: make refresh-uvx (as described in the README) or explicit uvx cache refresh commands.
• Common commands in usage include: guru_pk_help, export_session, export_session_as_infographic, export_enhanced_session, set_language.
• When updating guru-pk-mcp, use UVX to reinstall from PyPI or from your local source path as indicated in your setup instructions.
• This MCP setup emphasizes dynamic expert generation, multi-round debates, and Tufte-style visualizations; ensure your client supports the required output formats for the infographic exports.