MCP Technical Documentation

Welcome to the MCP Technical Documentation: a practical guide to connecting open public data to AI models using the Model Context Protocol (MCP).

This manual provides the architecture specifications, code patterns and configuration templates needed to build plugins, run the server and deploy data tools.

Everything here is measured against two goals: accuracy (answers computed from official data, not recalled from training) and traceability (every answer links back to its source). The project context explains both and why they matter for public data.

Looking for non-technical context or project strategy? Check out the Field Guide to Connecting AI to Public Information. It covers lessons from our Brazil and Uruguay pilots, guidance on working with domain experts, and real-world user feedback. The link to the Field Guide will be shared soon.

For more information, please visit the official project page "Traceable AI Answers for Public Data" at the Open Knowledge Foundation (OKFN) website.

What is in the box

• An MCP server that turns open datasets (CSV files, databases) into tools an AI can call.
• A chat gateway, a simple web chat that connects any OpenAI-compatible LLM to the MCP server and renders tables, charts and source links straight from the data, without routing them through the AI.
• Plugins scoped to a focused data domain (Uruguay's energy balance, Brasil, and yours next) that describe datasets and the questions they can answer.

Where to go next

• New here? Start with the project context.
• Want to run it? Go to getting started.
• Want to add your country's data? Read plugins.

Early days

The whole platform is in an early research phase. Breaking changes are expected, and so is this documentation changing under your feet.