MCP Knowledge Tools

This repository implements a Model Context Protocol (MCP) server that exposes three HTTP-backed retrieval tools for internal knowledge bases.

Available tools

Each tool sends a POST request to the Dify dataset retrieval API:

POST https://api.dify.ai/v1/datasets/{data_id}/retrieve
Authorization: Bearer <dataset token>
Content-Type: application/json
{"query": "<your question>"}

Configuration

Set the dataset token through the DIFY_DATASET_TOKEN environment variable. A fallback token (dataset-gCRaKZgnKtvqLdeuoCFjKiME) is bundled for quick testing, but production deployments should override it.

export DIFY_DATASET_TOKEN="dataset-..."

Installation

The project is published as a standard Python package and can be installed with your preferred Python packaging tool.

Using uv

uv offers fast Python environment management and package installation.

uv venv
source .venv/bin/activate
uv pip install .

Using pip

The project is published as a standard Python package.

python -m venv .venv
source .venv/bin/activate
pip install .

Running the MCP server

The pyproject.toml file defines a console script entry point. After installation you can launch the server via:

mcp-knowledge-server

Alternatively you can run the module directly:

uv run mcp-knowledge-server

The server uses FastMCP from the Model Context Protocol reference implementation, so it can be registered with compatible MCP clients.