UNDP Open Data MCP Server
This server provides tools to query and analyze UNDP's open data on development projects, funding, and donors via MCP.
Reference data: List operating units (country offices), regions, SDGs, SDG targets, focus areas, signature solutions, policy markers/approaches, and CRS sector codes.
Project search & details: Search projects by year, country, SDG, sector, donor, marker, etc.; fetch full project records with outputs, locations, and documents; retrieve individual output financials and CRS/gender markers.
Portfolio retrieval: Get a country office's entire project portfolio with pagination and optional outputs/locations.
Aggregation: Summarize a full year's funding by country, region, donor country, donor type, CRS sector, or core vs non-core — ideal for "which countries got the most funding" or "how much did Germany contribute" questions.
Donor intelligence: Search 6,200+ funding organizations by name/country; list donor countries ranked by contribution with contributing organizations.
Flexible filtering: Tools accept iso3 codes, years, budget sources, marker types/ids, and include flags for verbose data (outputs, documents, top donors).
Pagination support: Most list tools return
total,has_more, andnext_offsetfor paging through large result sets.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@UNDP Open Data MCP Server Which countries received the most UNDP funding in 2023?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
UNDP Open Data MCP Server
An MCP server for the UNDP Open Data API — UNDP's transparency data on projects, outputs, funding, donors, SDG alignment and focus areas.
No API key or authentication required.
Install
Sharing this with non-technical colleagues? See INSTALL.md for the one-file Claude Desktop extension and what to tell people about the data.
dist/ is gitignored, so build after cloning:
npm install && npm run buildRelated MCP server: @cyanheads/socrata-mcp-server
Running it
You do not start this server yourself — the MCP client launches it on demand and talks to it over stdio. To check it works standalone:
npm run build && node dist/index.jsIt should print UNDP Open Data MCP server running on stdio to stderr and then wait
for JSON-RPC on stdin. Ctrl-C to stop. Nothing further will happen without a client
attached; that is expected.
To exercise every tool against the live API:
node test/smoke.mjsConnecting to Claude
Claude Code (CLI)
From an interactive terminal:
claude mcp add undp-open-data --scope user -- node /absolute/path/to/undp-open-data-mcp/dist/index.js--scope user makes it available in every project. Use --scope project instead to
write a .mcp.json into the current project and share it with collaborators. Verify
with claude mcp list, or /mcp inside a session.
Claude Desktop
Edit ~/Library/Application Support/Claude/claude_desktop_config.json and add:
{
"mcpServers": {
"undp-open-data": {
"command": "node",
"args": ["/absolute/path/to/undp-open-data-mcp/dist/index.js"]
}
}
}Restart Claude Desktop, then check the tools appear in the MCP/tools menu.
Both need the absolute path, and node must be on the launching process's PATH.
If the client reports the server failing to start, run the node dist/index.js
command above by hand — the error will be clearer there.
Asking for things once connected
"Which countries received the most UNDP funding in 2023?" →
undp_aggregate_projects"What is UNDP doing in Kenya on climate?" →
undp_list_operating_units, thenundp_search_projects"How much has Germany contributed?" →
undp_list_donor_countries/undp_search_donors
Tools
Reference data
Tool | Purpose |
| Country offices and regional units. Start here to resolve a country name to its iso3 code. |
| The regional bureaux (RBA, RBAP, RBEC, RBLAC, RBAS) with aggregates. |
| The 17 SDGs with UNDP funding, filterable by year/unit/donor. |
| Target-level (1.1, 1.2 …) funding detail. |
| 'Our Focus' thematic pillars. |
| The six signature solutions. |
| Policy markers — supplies the |
| OECD DAC CRS sector codes. |
Projects
Tool | Purpose |
| Search projects for a year, filtered by country, SDG, sector, donor, marker. Paginated. |
| One project in full: dates, fiscal-year financials, outputs, locations, documents. |
| One output: CRS sector, gender marker, per-donor and per-year financials. |
| A country office's whole portfolio, paginated. |
| Totals across a full year grouped by country, region, donor country, donor type, CRS sector, or core vs non-core. |
Donors
Tool | Purpose |
| Search 6,200+ funding organisations by name. Query required. |
| Donor countries ranked by contribution, with contributing organisations. |
Design notes
The upstream API is generous with payload size — several endpoints return far more than fits in a model's context window:
Endpoint | Raw size |
| ~37 MB (6,221 donors) |
| ~7 MB (4,619 projects for 2023) |
| ~600 KB for a single country |
So no tool returns a raw upstream document. Instead:
Everything is paginated. Lists return
total,has_moreandnext_offset; passnext_offsetback asoffsetto continue.Verbose nested data is opt-in. Outputs, documents, locations, and
top_donorsarrays sit behindinclude_*flags, because they dominate the payload when left in.Text search covers the whole result set.
queryonundp_search_projectspulls every project matching the structured filters (pages fetched concurrently, ~2.5 s cold for a full year, then cached) and matches title and description across all of them, so counts are real rather than whatever landed on the first page.Aggregation happens server-side.
undp_aggregate_projectsreduces a whole year of projects to ranked totals, so "which countries received the most funding in 2023" costs one call instead of paging through 4,619 records.Large documents are cached in memory (1 h TTL by default), with concurrent requests for the same URL deduplicated. The donor index takes ~7 s cold and ~3 ms warm.
A hard size guard (100 k chars) backstops every result, returning a recoverable error with guidance rather than flooding the context.
Upstream quirks handled:
crs-index.jsoninverts its fields —idholds the label andnamethe numeric code. The server normalises this to{code, name}.project_listsupports undocumentedlimit/offsetparameters (visible only in itsnextlink); the server uses them for real server-side pagination.Output
budget,expenditureandfiscal_yeararrive as parallel arrays;undp_get_outputzips them into per-year records.sdg-indexandfocus-area-indexaccept anoperating_unitparameter (it is in the published spec) but return an empty array for every country, including ones with active projects. The server detects this and returns an explicit warning naming the working alternative, rather than an empty list that reads as "no such work in that country".In
group_by: "donor_country", per-donor amounts are used rather than the project total, so multi-donor projects aren't counted several times over. Note that upstream includes non-country funders (e.g.GFATM,UNDP) in the donor-country field.
The API's download/undp-project-data.zip endpoint is deliberately not exposed — it is a bulk
binary download, not something to hand back through a tool call.
Configuration
Variable | Default | Purpose |
|
| API base URL. |
|
| Cache lifetime. |
|
| Per-request timeout (the big endpoints are slow). |
|
| Tool result size ceiling. |
Tests
node test/schema.mjs # offline: schema contract, no network
node test/smoke.mjs # hits the live UNDP API
node test/minimal-args.mjs # hits the live UNDP APIsmoke.mjs drives the server as a real MCP client over stdio and exercises all 15 tools.
minimal-args.mjs calls every tool with optional parameters omitted, and fails if any input
schema emits a JSON-Schema default — Zod's .default() produced schemas that some MCP
clients rejected as required-but-missing, so defaults are applied in the handlers instead.
test/edge.mjs covers the 37 MB donor path, error handling and pagination continuity.
Data source
Data from the UNDP Open Data API. Financial values are USD;
budget is allocated funding and expenditure is spent.
Hosting it (remote connector)
dist/index.js is the stdio server local clients run. dist/http-server.js serves the same
15 tools over Streamable HTTP so it can be hosted and added to claude.ai as a custom
connector — the only way to reach it from the web app or mobile, which cannot talk to local
servers. See DEPLOY.md.
npm run build && npm run start:http # POST /mcp, GET /healthCI
GitHub Actions runs on every push and PR, plus weekly to catch upstream API changes:
Build and schema contract — Node 20 and 22.
test/schema.mjsinspectstools/listonly and does no network I/O, so it stays meaningful when the UNDP API is down.Live API tests — kept in a separate job so an upstream outage reads as an upstream problem, not a broken build.
Extension builds — packs the
.mcpb, runs the server from inside it, and uploads it as a build artifact.
test/as-desktop.mjs is excluded from CI: it reads the local claude_desktop_config.json,
which does not exist on a runner.
Disclaimer
An independent project that reads UNDP's public Open Data API. Not an official UNDP product and not endorsed by UNDP. Data belongs to UNDP; see api.open.undp.org for their terms.
Licensed MIT — see LICENSE.
Available Tools
15 toolsundp_aggregate_projectsAggregate UNDP project fundingA
Aggregate a full year of UNDP projects by operating unit, region, donor country, donor type or CRS sector, returning totalled budget, expenditure and project counts. This is the tool for questions like 'which countries got the most funding in 2023' or 'how much did Germany fund' — it summarises the whole year server-side instead of paging through thousands of projects.
Inputs: year (required), limit, region, sort_by, +3 more.
undp_get_operating_unitGet an operating unit's portfolioA
Get one operating unit's (country office's) full project portfolio, including each project's outputs and subnational locations. The upstream document is large, so projects are paginated and outputs are summarised by default.
Inputs: iso3 (required), limit, query, offset, +2 more.
undp_get_outputGet a UNDP project outputA
Get one project output: its CRS sector, gender marker, SDG tags, and budget / expenditure / disbursement broken down by fiscal year and donor. Output ids come from undp_get_project.
Inputs: output_id (required).
undp_get_projectGet a UNDP projectA
Get the full record for one UNDP project: dates, budget and expenditure by fiscal year, outputs, subnational locations and linked documents. Project ids look like '00122701' and come from undp_search_projects.
Inputs: project_id (required), include_outputs, include_documents, include_locations.
undp_list_approachesList UNDP approaches and markersA
List 'Our Approaches' — the policy markers UNDP tags projects with (gender, capacity development, and similar). The marker types and ids returned here feed the marker_type / marker_id filters on undp_search_projects.
Inputs: year, operating_unit, include_top_donors.
undp_list_crs_sectorsList OECD DAC CRS sector codesA
List the OECD DAC CRS sector codes used to classify UNDP outputs (e.g. 15162 'Human rights'). Useful for decoding the crs field on an output.
Inputs: query.
undp_list_donor_countriesList UNDP donor countriesA
List countries and territories that contribute funding to UNDP, with total budget and the contributing organisations behind each. Use query to look one up, or list them ranked by contribution.
Inputs: limit, query, offset, sort_by, +1 more.
undp_list_focus_areasList UNDP focus areasB
List UNDP's focus areas ('Our Focus' thematic pillars) with budget, expenditure, project and donor counts. Optionally include each area's top donors.
Inputs: year, budget_source, operating_unit, include_top_donors.
undp_list_operating_unitsList UNDP operating unitsA
List UNDP operating units (country offices and regional/global units) with headline budget, expenditure and project counts. Use this to resolve a country name to the iso3 code that every other tool's operating_unit filter expects.
Inputs: limit, query, offset, sort_by, +1 more.
undp_list_regionsList UNDP regionsA
List the UNDP regional bureaux (RBA, RBAP, RBEC, RBLAC, RBAS, and global units) with aggregate budget, expenditure and project counts, optionally including the countries in each region.
Inputs: include_countries.
undp_list_sdgsList SDGs with UNDP fundingA
List the 17 Sustainable Development Goals with UNDP project counts, budget and expenditure. Filterable by year, operating unit and budget source. Set include_targets for target-level detail on a single SDG.
Inputs: sdg, year, budget_source, operating_unit, +1 more.
undp_list_sdg_targetsList SDG targetsA
List SDG target-level data (the 1.1, 1.2 … sub-goals) with associated UNDP funding. Filter by SDG code, target id and year.
Inputs: sdg, year, limit, offset, +2 more.
undp_list_signature_solutionsList UNDP signature solutionsA
List UNDP's six signature solutions (the Strategic Plan delivery areas) with aggregate funding and project counts.
Inputs: include_details.
undp_search_donorsSearch UNDP donorsA
Search the index of organisations that fund UNDP (governments, multilaterals, foundations, private sector) by name or id. A query is required — the full index holds thousands of donors and is too large to list. Returns the donor id used as budget_source elsewhere.
Inputs: query (required), limit, offset, country.
undp_search_projectsSearch UNDP projectsA
Search UNDP projects for a given year, filtered by country, SDG, focus area, signature solution, budget source or policy marker. Returns a compact record per project (title, budget, expenditure, SDGs, donors). Results are paginated — follow next_offset for more. Use undp_get_project for the full detail of one project.
Inputs: year (required), sdg, limit, query, +10 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 15 tools
Most tools are clearly distinct by resource type (operating units, regions, SDGs, projects, donors), but undp_list_operating_units and undp_get_operating_unit could be confused at first glance, and undp_list_sdgs vs undp_list_sdg_targets are similar in name though distinct in granularity. Overall, descriptions clarify boundaries well.
The set follows a consistent undp_verb_noun pattern: list_*, get_*, search_*, aggregate_*. Minor deviation: undp_list_donor_countries is a list but requires a query, and undp_search_donors vs undp_list_donor_countries have overlapping list/search semantics. Still, the pattern is predictable.
15 tools is well within the ideal range for a data-rich open data server. Each tool covers a distinct facet of the UNDP data model (reference lists, project search/detail, aggregation, donor lookup), and none feel redundant or superfluous.
The surface covers the core domain well: reference data (regions, SDGs, focus areas, sectors), project search and detail, output detail, operating unit portfolios, and aggregate queries. Minor gaps: no direct tool for listing all projects without search filters, and no tool for fetching a specific donor's full profile, but these are workable via existing tools.
Maintenance
Related MCP Connectors
Search and preview datasets publicly. Account-required full data via MCP.
Query UNHCR refugee, IDP, and stateless populations, asylum decisions, returns, and resettlement.
Free, read-only public-source threat intel over MCP. No account or API key.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceEnables querying international public procurement data, including government tenders, through the Open Contracting standard.90 npmMIT
- AlicenseNot gradedqualityBmaintenanceSearch and query government open-data portals (Socrata SODA API) via MCP.592 npm4Apache 2.0
- AlicenseNot gradedqualityCmaintenanceEnables querying Australian government tender data via the AusTender OCDS API without requiring an API key.91 npmMIT
- AlicenseNot gradedqualityDmaintenanceProvides a unified interface to access UNHCR's open data across statistics, RDF, and IATI MCP servers, enabling aggregated queries, cross-domain analytics, and dataset discovery.MIT