UN System Data Commons MCP server
This book is a practical entry point for anyone who has a basic understanding of the Model Context Protocol (MCP) and wants to start exploring the UN System Data Commons MCP server.
The current access point is:
https://unsd-datacommons.gcp.un-icc.cloud/mcpThe aim is not to explain MCP in depth. It is to show how to confirm that the server is reachable, inspect what it exposes, understand the main research workflows, and make a first useful query.
The mental model¶
The MCP server is a machine-oriented interface that lets MCP-capable clients work with the UN System Data Commons in a structured way: tools that search its statistical corpus and retrieve observations, and resources that carry the research playbooks an AI client is expected to follow.
How this book is organised¶
What the server does — The tools, the resources, why the playbooks matter, and the statistical and geographic scope of this deployment.
Getting started with the Inspector — Confirm the server is reachable, add it to the MCP Inspector, and run a first request from the command line or your own client.
Tools and playbooks — The six tools, the three
SKILL.mdplaybooks and the operating rules they set.Research patterns — A recommended first exploration, the three research patterns, and a first-session checklist.
Inspecting the graph with the REST API — Five structural questions the graph’s node endpoint answers around identifiers the MCP has established, and the boundary rule that keeps them UN-scoped.
Concepts — DCIDs, statistical variables, places, provenance.
Playbooks — The three
SKILL.mdresources, rendered verbatim from the server: general research, child places, multi-entity.Walkthroughs — The research patterns run for real against the live server at build time: A, an indicator for one place, B, an indicator across child places and C, relationships between entities — which finds that this instance has no multi-entity variables yet, and shows what to do instead.