mcp-openfoodtox

This MCP server provides natural language access to EFSA’s OpenFoodTox chemical hazards database. It exposes a set of tools that let you search for substances, retrieve safety assessments, toxicity endpoints, risk assessments (ADI/TDI), genotoxicity details, EFSA opinions, and filtered lists of substances by class or assessment. You can interact with the server through natural language prompts such as: “What is substance X?”, “Is Y safe?”, “What are the toxicity effects of Z?”, “List substances with ADI above 5 mg/kg.” The bundled tools translate these queries into structured data pulled from the EFSA dataset, returning concise safety flags, study results, and regulatory information. To use it, run the server with your Python environment, then issue queries via Claude Desktop or your preferred client that supports MCP endpoints, using the described prompts to access the different capabilities.

Prerequisites:

• Python 3.12+ installed on your system
• uv package manager installed (see prerequisites section in README)
• Access to Claude Desktop or an MCP client that supports MCP endpoints

Installation steps:

1. Install uv (Python environment manager):

curl -LsSf https://astral.sh/uv/install.sh | sh

Optionally, install via Homebrew (macOS):

brew install uv

2. Ensure Python 3.12+ is available via uv (or system Python):

uv python install 3.12

3. Set up and run the MCP server locally (first run requires database setup):

# From the repository root of mcp-openfoodtox
make setup

No virtual environment activation is required for the quick start; the setup handles dependencies. 4) If you prefer manual Claude Desktop configuration, use:

# Example manual config (adjust paths accordingly)
# absolute/path/to/mcp-openfoodtox/.venv/bin/python absolute/path/to/mcp-openfoodtox/main.py

5. Start the server and verify it responds to queries via your MCP client (Claude Desktop or another compatible tool).

Note: The project relies on the OpenFoodTox dataset updated September 14, 2023. Ensure you have network access for data loading and that prerequisites like Python 3.12+ are properly installed.

Tips and notes:

• The server is Python-based and is designed to be run from a virtual environment created within the mcp-openfoodtox folder (as indicated by the manual config example).
• If you run into issues with Claude Desktop integration, verify the config.json path and ensure the specified Python executable and main.py are accessible.
• The dataset originates from EFSA OpenFoodTox; for regulatory use, always cross-check with the original EFSA outputs.
• To switch between local MCP server runs or to adjust resource limits, consider environment variables or containerization (not required but supported by alternative deployment methods).
• If you are behind a proxy or have restricted network access, ensure your environment can reach the EFSA OpenFoodTox endpoints and any data download requirements.
• No npm package is required for this server (npm_package is null).