prolog-mcp-server

prolog-mcp-serverは、Prologプログラムの実行と管理を支援するPythonベースのサーバーです。ユーザーは、Prologのクエリを送信し、結果を受け取ることができます。このツールは、AIや自動化のワークフローにおいて、Prologの強力な論理推論能力を活用するための便利なインターフェースを提供します。

GitHubスター

3

ユーザー評価

未評価

お気に入り

0

閲覧数

13

フォーク

1

イシュー

0

README
Prolog MCP Server

A Model Context Protocol (MCP) server that enables direct integration between Large Language Models and SWI-Prolog. This server allows LLMs to define Prolog facts and rules (clauses) and to execute Prolog queries and leverage logical reasoning capabilities.

Features
  • Allow LLMs to add facts and rules (clauses) to SWI Prolog server
  • Execute Prolog queries through MCP protocol
  • Support for both HTTP and stdio transport
  • Docker containerization support for the SWI Prolog server
  • Implemented as python package for easy integration to agentic LLM applications
  • Compatible with MCP Inspector for interactive testing and debugging

MCP Inspector Interface
Prolog MCP Server running in MCP Inspector showing available tools and interface

Prerequisites
  • Python 3.8 or higher
  • Git (for installation from source)
  • Docker
Installation on linux
  1. Make sure git, python-venv and docker are installed, i.e.
    Install git
sudo apt install git python3.12-venv docker-buildx
  1. Clone git repo
git clone https://github.com/wendelinism/prolog-mcp-server.git

Move into newly downloaded folder

cd prolog-mcp-server/
  1. Create local python environment
python3 -m venv .

Activate local python environment

source bin/activate
  1. Install python package prolog-mcp-server, and all it's python dependencies
pip install -e .
  1. Build prolog docker image (without this the MCP-server will have no prolog server to connect to)
docker buildx build -t prolog-docker-image -f docker/prolog.dockerfile .

In case your user does not have permission yet to interact with docker, add local user to docker group:

sudo usermod -aG docker $USER

To make this change effective, restart the terminal session with

exec su - $USER

then go back to the project folder. Now you should be able to build the docker image.

Installation on Windows and Mac

Should work on windows and Mac OS as well, but not tested yet.

Usage
Integration with Claude Desktop

Claude Desktop with Prolog-MCP-server connector
Claude Desktop with connector to Prolog-MCP-server

Claude Desktop with avalable tools for Prolog-MCP-server
Claude Desktop can choose to use any of the tools shown for connected Prolog-MCP-server

If Claude Desktop is installed, the Prolog MCP server can be added to the Claude config, so that Claude can start a local instance of the server and run prolog code directly.
Preconditions:

  • Local clone of prolog-mcp-server repository (see above)
  • Local python environment installed (see above)
  • Docker prolog-docker-image built (see above)
Configuration

Add the following configuration to your claude_desktop_config.json file:

Linux/Mac: ~/.config/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "prolog-mcp-server": {
      "command": "/path/to/python-env/bin/python",
      "args": [
        "/path/to/prolog-mcp-server/src/prolog_mcp_server.py",
        "--transport=stdio"
      ],
      "env": {
        "PATH": "/path/to/python-env/bin:/usr/local/bin:/usr/bin:/bin"
      }
    }
  }
}

Note: Update both /path/to/python-env/ and /path/to/prolog-mcp-server/ to match your actual installation paths.

Starting the Standalone Server with HTTP

For HTTP transport setup, see the the minimal example implemenation to start the standalone Prolg MCP server at HTTP transport implementation.

Examples for HTTP and stdio usage

Check the examples/ directory for:

To start standalone Prolog MCP server with http server:

python3 examples/demo_prolog_MCP-server-start_http.py
Test MCP-server with MCP inspector

To test a running MCP server with MCP inspector, first install npm:

sudo apt install npm -y

When MCP server is running with http transport, run in separate terminal:

fastmcp dev examples/demo_prolog_MCP-server-start_http.py

Follow link to MCP inspector with prefilled token.
In the web interface make sure to adjust transport type ("Streamable HTTP") and URL ("http://localhost:8000/mcp")
Hit connect, for test you should be able to go to "tools" now, and there "list tools"

Docker Support

A Dockerfile is provided in the docker/ directory for containerized deployment.

License

MIT License - see LICENSE file for details.

Contributing

Contributions are welcome. Feel free to submit issues and pull requests.

Credits

This MCP server relies on SWI-Prolog, an open-source Prolog implementation. All credits for the actual Prolog implementation go to the SWI-Prolog development team.