What the server does - UN System Data Commons MCP server

MCP provides a standard way for an AI client or development tool to discover and call capabilities exposed by a server. For the UN System Data Commons, the MCP server currently exposes tools for:

Searching for statistical indicators published through the UN System Data Commons;
Inspecting metadata for selected statistical variables;
Retrieving observations for a place;
Discovering and retrieving indicators across the places contained in a parent geography, such as the countries within a region;
Retrieving observations across multiple related entities, such as bilateral or flow-type data.

The server also exposes resources containing research playbooks. These are important because they explain how an AI client is expected to use the tools.

Statistical scope¶

Important

The UN System Data Commons MCP is intended for statistical research using content published through the UN System Data Commons: its statistical variables, their observations and their provenance. The knowledge graph behind it also uses shared Data Commons reference entities — places, units, classifications, enumeration values, semantic concepts — wherever common identity is useful. A shared reference entity does not widen the statistical scope: a country node held in common with the wider graph provides shared geographic identity; it does not by itself expand the statistical corpus being researched. The boundary that matters is the statistical content selected and presented.

The UN System Data Commons harmonizes what the agencies’ data have in common: the concepts and classifications used as breakdowns, such as sex, age and urbanization; units; place identifiers; and the thematic organization the topics express. That shared vocabulary is what makes variables from many UN agencies discoverable together and comparable in structure. It does not make source series or agency methodologies interchangeable: each variable remains its agency’s series, produced by that agency’s method, and similar variables from different agencies are distinct variables. Three agencies publish a female life expectancy at birth; those are three variables, not three sources of one.

In normal research, discover statistical variables with the search tools and carry the identifiers they return into metadata and observation retrieval:

research question
    ↓
search_indicators / search_child_indicators
    ↓
candidate statistical variables from the UN System Data Commons
    ↓
get_variable_metadata
    ↓
the selected candidate
    ↓
get_observations / get_child_observations

In normal research workflows, retrieval should operate on a statistical-variable DCID produced by the preceding discovery step. A DCID is an identifier, not evidence that an entity belongs to this corpus, and the retrieval tools take the DCID they are given as is; so do not substitute variable DCIDs obtained from another Data Commons environment or from generic examples. The research rules put this in six lines.

Geographic scope of the current data¶

Important

The current UN System Data Commons publishes statistical observations for countries and for supranational geographies: the world, continents, UN regions such as Sub-Saharan Africa, and other geographic groupings such as the Small Island Developing States. It does not currently publish subnational observations.

Places below country level — states, provinces, districts, cities — may nevertheless exist and resolve in the underlying knowledge graph. A search scoped to them succeeds and returns no variables. Read that as the platform’s current geographic coverage, not as evidence that the statistical concept is absent; the same concept is usually available at country level.

The parent-and-child tools are therefore used today primarily to navigate supranational geographies and retrieve or compare country-level observations within them: the countries within the world, a continent, a UN region or a grouping. This concerns the data currently loaded, not the tools, which traverse whatever levels of the place hierarchy carry data.

A value returned for a region, a continent or a grouping is an observation the source agency published for that geography. The UN System Data Commons does not compute it by aggregating the country observations a child-place query returns, and the countries the graph records as contained in a grouping need not be the ones the source aggregated.

The server, as it introduces itself¶

Everything below this card is regenerated from the live server each time the site is built, so it is the server’s current identity (not a copy of it).

	Reported by the server
Server	DC MCP Server
Version	1.3.0
Protocol	2025-06-18
Capabilities	experimental, extensions, logging, prompts, resources, tools

The server’s own instructions

Act as a Data Commons Research Assistant. This server provides direct access to a massive, unified knowledge graph of aggregated statistical data from authoritative regional and global sources like the UN, World Bank, and Census Bureau. Use it to transform natural language queries into precise statistical insights by identifying specific indicators and retrieving their observations. It contains historical and recent data points on topics like demographics, economics, health, and environment across various geographic levels. It does not contain information on topics like real-time news, subjective viewpoints, or private corporate data.

CRITICAL INSTRUCTION: When performing statistical research, searching for indicators, or fetching observations, you MUST first read the appropriate MCP Resource before calling any tools. Use your platform’s standard MCP resource-reading capability to retrieve:

skill://data-commons-researcher/SKILL.md (for single-place queries)
skill://data-commons-child-places-researcher/SKILL.md (for child-places, sub-national breakdowns, or geographic hierarchies)
skill://data-commons-multi-entity-researcher/SKILL.md (for bilateral relationships, flows, foreign aid, trade, or multi-entity queries)

Crucially, every data point retrieved must be attributed to its original source provided in the tool output; never present statistics as “known facts” without citing the specific organization or dataset they originated from. Prioritize data integrity and transparency, ensuring that users understand both the metric and the provenance of the information provided.

CRITICAL RULE ON DCIDs: Before calling any observation or metadata tool, you MUST resolve variable and place DCIDs using search tools (search_indicators or search_child_indicators) even if it means making a similar looking search call again for a place or entity you don’t have the DCID for. DO NOT guess or hardcode DCIDs under ANY circumstances. Never assume a DCID based on similar looking DCIDs you have seen previously.

A capability in that table means the server implements that part of the protocol, not that it has content for it: prompts is listed because the server can list prompts and will notify clients when that list changes, and it currently publishes none.

The instructions in the card are likewise the upstream Data Commons server’s generic text, and two phrases in them are broader than this deployment: the sources here are UN system agencies (the prefixes Walkthrough C lists), not the Census Bureau, and the sub-national breakdowns the child-places skill is offered for are not loaded on this instance — see Statistical scope and Geographic scope above.

A word on names. This book says playbook for the three research documents. The server namespaces them as skills (skill://…) and delivers them through the MCP resources mechanism; all three words point at the same files.

Tools, resources, skills, prompts¶

A useful distinction when first looking at the server:

Tools: Callable operations. They search the knowledge graph, inspect metadata, or retrieve observations. The server currently provides six.
Resources: Server-provided documents or other readable content, addressed by URI.
Skills: Operational playbooks exposed as resources under skill://…/SKILL.md. They tell an AI client how to choose and sequence the tools. There are three, one per research pattern, each with a _manifest resource listing its files.
Prompts: Another MCP capability, but the server currently exposes none — prompts/list returns an empty list, and that is normal. The main interface is through tools and resources.

The three research patterns¶

Skill	When to use it	Main routing guidance
`data-commons-researcher`	Normal statistical research for one or more specified places	Discover candidate indicators, assess metadata and provenance, then retrieve observations
`data-commons-child-places-researcher`	Questions about statistics across the places contained within a parent geography — today, the countries within the world, a continent, a UN region or another geographic grouping	Switch to this skill when the question spans the countries within a parent geography
`data-commons-multi-entity-researcher`	Statistical variables that describe relationships between multiple entities	Switch to this skill when metadata shows multiple observation entity properties, such as donor/recipient

For most users the place to start is the general playbook, skill://data-commons-researcher/SKILL.md. It defines a three-step pipeline — discovery with search_indicators, assessment with get_variable_metadata, retrieval with get_observations — and two routing rules that hand off to the other two skills.

Where the details live¶

The tool schemas visible in the MCP Inspector, and the SKILL.md resources themselves, are the best source for the exact current parameters and rules. You do not need to memorise them — the next page shows how to fetch them.