mcp-server-agribalyse

mcp-server-agribalyse

🌿 A Model Context Protocol (MCP) server for querying the ADEME Agribalyse 3.1 dataset.

This MCP server provides tools and resources to interact with the Agribalyse public API, enabling Large Language Models to retrieve and analyze environmental impact data on food products.

⚠️ Note: This server is based on the FastMCP framework and is actively maintained. API coverage may evolve.

---

🚀 Overview

Agribalyse is a dataset published by ADEME, offering environmental indicators (climate impact, water use, ecotoxicity, etc.) for thousands of food products.

This MCP server allows LLMs to:

• Search food product data.
• Aggregate environmental metrics.
• List possible filter values.
• Understand dataset structure.

---

🧰 Tools

Tool Name	Description
read_lines	Query rows from the Agribalyse dataset
get_values	Get distinct values for a given text field
get_metric_agg	Compute a single metric (avg, min, etc.) on a numeric field
get_simple_metrics_agg	Compute metrics on one or more fields at once
get_words_agg	Retrieve most frequent tokens in a text field
read_schema	Get the complete column schema of the dataset
read_safe_schema	Get a reduced version of the column schema
read_api_docs	Fetch the full OpenAPI specification from the ADEME API

---

📚 Resources

URI	Description
agribalyse://fields	List of allowed text fields for querying
agribalyse://files	Available data files published by ADEME
agribalyse://sample-lines	A sample of dataset rows
agribalyse://columns/sortables	Fields that can be used for sorting
agribalyse://metrics/fields	Numeric fields usable for aggregation
agribalyse://metrics/types	Supported metric types (avg, sum, percentiles, etc.)
agribalyse://fields/descriptions	Human-readable descriptions of each dataset column

---

💬 Prompts

This server also includes predefined prompts for easier interaction:

• search_product: Ask for environmental info about a named product
• ask_stat: Ask for a specific metric on an indicator
• compare_products: Compare two products by one indicator
• list_field_values: List possible values of a given field
• sample_prompt: Ask to preview sample data
• explain_indicator: Ask for an explanation of an indicator
• custom_query_prompt: Prompt chain for guided query refinement

---

🧪 Debugging

You can inspect server behavior using:

npx @modelcontextprotocol/inspector uvx run src/agribalyse/server.py

Or follow logs using:

mcp dev server.py

🧑‍💻 Client Example

To test the MCP server using a client, an example client implementation using OpenAI is provided.

Run the client with:

python client/client.py

This will execute example prompts and display the server responses, allowing you to observe how the MCP server handles requests.

🧪 Running Tests

To run the test suite using pytest, make sure your virtual environment is activated and then run:

pytest tests/test_mcp_tools

This will execute all unit tests and validate the MCP tools integration.

👩‍💻 Maintainer

Author: Tracy André

Organization: Positive Solutions

Contact: [email protected]