Sdmx Structure MCP Server

@pipeworx/sdmx-structure

Dataflow dimensions and code lists — "what do the codes in this dataset mean" — for any SDMX 2.1 statistical registry: Eurostat, OECD, the European Central Bank, the UN Statistics Division (via the data.un.org SDMX hub), the IMF, the Bank for International Settlements, ILOSTAT, and the Australian Bureau of Statistics. This is the STRUCTURE half of SDMX, distinct from pulling observations — use it to decode a dimension's codes or to learn the key order a registry's own get_data tool needs.

Part of Pipeworx — an MCP gateway connecting AI agents to 1784+ live data sources. This is an independent, unofficial integration — not affiliated with, endorsed by, or published by the upstream provider.

Tools

• sdmx_dataflow_structure(registry, dataflow, agency?, version?, max_codes_per_dimension?) — the ordered dimensions of one dataflow, each with its codelist's code→label pairs (capped per dimension, with a truncation flag and the true total count). Resolves the dataflow's real DSD (Data Structure Definition) reference rather than assuming the dataflow id equals the DSD id — they differ for many flows (Eurostat's UNE_RT_M dataflow v1.0 points at DSD UNE_RT_M v155.0; BIS's WS_CBPOL dataflow points at a differently named DSD, BIS_CBPOL).
• sdmx_codelist(registry, id, agency?, version?, filter?, limit?) — one codelist's code→label pairs by id alone, with no dataflow needed. Useful once you already know a codelist id (from sdmx_dataflow_structure's codelist field, or from documentation) and just want the codes, optionally filtered by a substring.
• sdmx_find_dataflow(registry, query, agency?, limit?) — keyword search across a registry's dataflow catalogue by id or name, to find the id to pass to sdmx_dataflow_structure.

Registries

OECD, IMF and UNSD default to agency all, not a single agency — this is load-bearing, not a placeholder. All three registries split their catalogue across many maintaining sub-agencies and there is no single agency id that covers it:

• OECD: there is no agency literally called "OECD" — every dataflow is published under a directorate (OECD.SDD.NAD, OECD.ELS.JAI, OECD.EDU.IMEP, ...; 52 distinct agency ids observed live, some mirroring in ESTAT and IAEG-SDGs flows too). Querying agency "OECD" 404s with "No Results Found" on both the catalogue listing and a single-dataflow lookup — this is the live defect fleet #2839's follow-up caught (sdmx_find_dataflow({registry:"oecd", query:"unemployment"}) 404ing).
• IMF: ten sub-agencies (IMF.STA, IMF.MCD, IMF.SPR, IMF.AFR, IMF.MCM, IMF.FAD, IMF.WHD, IMF.APD, IMF.RES, plus the external ISORA). IMF.STA alone 200s — which is why this looked fine at first — but covers only 190 of 223 dataflows; the other 33 are every regional economic outlook (APDREO, AFRREO, WHDREO, ...) and the fiscal affairs dataflows.
• UNSD: the hub mirrors ESTAT, IAEG-SDGs, UIS and World Bank dataflows under THEIR agency ids alongside UNSD's own ~4. Querying agency "UNSD" 200s but returns only UNSD's own flows (1.4KB observed) where "all" returns the whole hub (4.4KB, ~15 dataflows).

"all" resolves a SINGLE dataflow lookup identically to the dataflow's real sub-agency for all three (verified side by side, byte-identical response, for an OECD.SDD.NAD flow, an IMF.APD flow and an ESTAT-owned UNSD-hub flow) — so there is no coverage cost to defaulting to it. Pass agency only to narrow a search to one sub-agency, or to match a different registry's single-agency default (Eurostat/ECB/BIS/ILOSTAT/ABS, each already its own one agency).

Not included: Stats NZ. Its SDMX endpoint requires a platform API key (Ocp-Apim-Subscription-Key, injected by the gateway from PLATFORM_STATSNZ_KEY) and a literal, non-URL-encoded comma in its flowRef path segment — neither fits this pack's keyless, URL-encoded-path design, and the stats-nz pack already ships a dedicated structure tool. See that pack.

Overlap with existing packs — read before calling this for ILO or ABS. The ilostat pack already ships dataflow_structure for ILO dataflows, and the abs-au pack already ships dataflow_structure for ABS dataflows — both inline codes directly, the same job sdmx_dataflow_structure here does. This pack supports registry: "ilo" and registry: "abs" only for the standalone sdmx_codelist lookup (a codelist by id, no dataflow needed) that neither of those packs exposes. Prefer ilostat's / abs-au's own dataflow_structure for an ILO or ABS dataflow — every response from this pack for those two registries repeats that in an overlap field.

Auth

Keyless. No auth on any of the eight registries above.

Data sources

• Eurostat SDMX 2.1 REST API — https://ec.europa.eu/eurostat/api/dissemination/sdmx/2.1 — structure endpoints (/dataflow, /datastructure, /codelist) answer SDMX-ML XML only; every JSON Accept header and format=SDMX-JSON query param this pack tried either 406s or is silently ignored (confirmed live 2026-10-08). The eurostat pack's own data endpoint (/dissemination/statistics/1.0/data) is a different API that does speak JSON-stat — the two are not interchangeable.
• OECD SDMX 2.1 REST API — https://sdmx.oecd.org/public/rest.
• European Central Bank Data Portal — https://data-api.ecb.europa.eu/service — structure endpoints are XML only, same split the ecb pack already documents for its own /dataflow call (fleet #1158): the DATA endpoints speak SDMX-JSON, the STRUCTURE endpoints (/dataflow, /datastructure, /codelist) do not.
• UN Statistics Division SDMX hub — https://data.un.org/legacy/ws/rest — the documented https://data.un.org/ws/rest/... path 302-redirects here; this pack calls the redirect target directly. It is a small, multi-agency hub (~15 dataflows observed) carrying UNSD's own flows (DF_UNDATA_ENERGY, DF_UNDATA_COUNTRYDATA) alongside mirrored IAEG-SDGs, UIS and World Bank flows under their own agency ids — pass agency if a dataflow you want belongs to one of those rather than UNSD.
• IMF SDMX 2.1 REST API — https://api.imf.org/external/sdmx/2.1. IMF concept-level representation trap: some IMF dimensions carry no localRepresentation.enumeration at all — e.g. EER's COUNTRY and INDICATOR dimensions. Their codelist is declared on the shared CONCEPT instead: the dimension's conceptIdentity URN points at a concept scheme (e.g. IMF:CS_MASTER_DATA), and that concept's own coreRepresentation.enumeration is the real codelist (IMF:CL_COUNTRY for COUNTRY). This pack follows that chain automatically; a naive implementation that only reads dimension.localRepresentation will report those dimensions as codeless.
• Bank for International Settlements SDMX 2.1 REST API — https://stats.bis.org/api/v2/structure (note the /structure segment — BIS nests dataflow/datastructure/codelist resources under it, unlike every other registry here).
• ILOSTAT SDMX 2.1 REST API — https://sdmx.ilo.org/rest. Accept-Language trap: ILO's backend returns a bare 500 languageTag1 body for any request with no Accept-Language header at all — reproduced with Node's/Workers' fetch (which sends none by default) but NOT with curl, confirmed live 2026-10-08. This pack sends Accept-Language: en on every outbound request to every registry (harmless elsewhere) specifically because of this.
• Australian Bureau of Statistics Data API — https://data.api.abs.gov.au/rest — the documented host api.data.abs.gov.au does not resolve; this is the same live host the abs-au pack already found and documented.

Size trap, all registries: fetching a DSD with references=children pulls in every codelist it uses in one response — Eurostat's UNE_RT_M alone is 3.7MB because its GEO (country/region) codelist is 2.6MB. This pack never does that: it fetches the DSD with references=none (a few KB — just the dimension list and each dimension's codelist REFERENCE), then fetches only the codelists actually needed, one at a time, each capped to max_codes_per_dimension (default 50, max 500) with a truncation flag and the true total code count.

A 404 on a structure endpoint always means "no such id" — unlike a data endpoint, there is no "dataflow exists but no observations matched" ambiguity for a structure query, so this pack does not need the data-endpoint no_records/no_dataflow split that shared/src/sdmx-miss.ts provides for other packs' get_data tools. Every 404 here comes back as a structured {error: "not_found", ...} naming sdmx_find_dataflow as the next step. A 200 that parses to zero dimensions or zero dataflows is reported as a loud upstream_parse: error, never as a silent empty result.

Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "sdmx-structure": {
      "url": "https://gateway.pipeworx.io/sdmx-structure/mcp"
    }
  }
}

What this endpoint actually serves

tools/list at https://gateway.pipeworx.io/sdmx-structure/mcp returns the tools in the table above plus the shared Pipeworx meta-tools — ask_pipeworx, discover_tools, search_within, remember/recall and the rest of the gateway-wide set. So the tool count you see is larger than this table: a single-pack endpoint currently lists roughly 30 shared tools alongside the pack's own. The connection's initialize response states its exact scope, and is the authoritative answer for a given day.

This is deliberate, not multiplexing by accident. The meta-tools are what let a scoped connection answer a question this pack does not cover — via ask_pipeworx, which routes across the whole catalog — without you adding a second MCP server. There is currently no way to mount a pack endpoint without them; if the extra schemas cost you more context than the routing is worth, connect to the full gateway once rather than to several pack endpoints.

Or connect to the full Pipeworx gateway to get every pack's tools listed directly, instead of just this one's:

{
  "mcpServers": {
    "pipeworx": {
      "url": "https://gateway.pipeworx.io/mcp"
    }
  }
}

Both URLs reach the same gateway and the same 1784+ data sources. The only difference is which pack's tools are listed directly; ask_pipeworx reaches all of them from either one.

No MCP client? Call it over HTTP

curl -X POST https://gateway.pipeworx.io/v1/tools/sdmx_dataflow_structure \
  -H 'Content-Type: application/json' \
  -d '{"registry":"eurostat","dataflow":"une_rt_m"}'

No account needed for the first calls. Inspect any tool: GET https://gateway.pipeworx.io/v1/tools/sdmx_dataflow_structure. Find one: POST https://gateway.pipeworx.io/v1/tools/search_packs with {"query":"..."}.

Standalone (no gateway account)

This package also runs as a local stdio MCP server — no Pipeworx account, no gateway round-trip:

{
  "mcpServers": {
    "sdmx-structure": {
      "command": "npx",
      "args": ["-y", "@pipeworx/mcp-sdmx-structure"]
    }
  }
}

Or run it directly to confirm it starts:

npx -y @pipeworx/mcp-sdmx-structure

It speaks MCP over stdin/stdout and answers initialize/tools/list/tools/call for only this pack's tools — none of the shared meta-tools the gateway connection above adds. Same source, same tools, no ask_pipeworx routing.

Using with ask_pipeworx

Instead of calling tools directly, you can ask questions in plain English — this works on the pack endpoint above as well as on the full gateway:

ask_pipeworx({ question: "your question about Sdmx Structure data" })

The gateway picks the right tool and fills the arguments automatically.

More

• Docs and guides
• pipeworx.io

License

MIT