ILO Labour Statistics (ILOSTAT): unemployment, wages, informality

ILO Labour Statistics (ILOSTAT) — MCP Server

[MCP] [CI] [Version] [Tools] [Resources] [Prompts] [npm] [MCP Registry] [ilo-mcp-server MCP server] [smithery badge] [License: MIT] [Status]

🇧🇷 Leia em Português

A public, hosted, provenance-first MCP server for the International Labour Organization (ILO) statistics — the ILOSTAT database — no installation, no account, no API key. Point your MCP client at the hosted endpoint and ask about unemployment, employment, wages, working time and other labour indicators by country, year, sex and age. It runs on Cloudflare Workers over Streamable HTTP and talks to the official ILOSTAT SDMX REST API.

Independent project. This is an unofficial, community-built client of the ILO's public ILOSTAT API — not affiliated with or endorsed by the International Labour Organization. Data remain © ILO under CC BY 4.0; see Data license and attribution.

Every response carries a provenance block (source URL, data vintage, real retrieval timestamp, license, ILO citation) — exact figures with an audit trail, not numbers guessed from training data.

🇧🇷 Em português. Servidor MCP remoto e hospedado (nada para instalar, sem conta e sem chave) para as estatísticas de mercado de trabalho da OIT — desemprego, emprego, salários, jornada e informalidade por país, ano, sexo e idade, direto no Claude, no ChatGPT ou em qualquer cliente MCP, com proveniência e citação da fonte em cada resposta: README em português.

Questions it answers

In plain language, inside the MCP client — the assistant picks the tool and the filters:

• "What has happened to unemployment in Brazil since 2015?" (ilo_get_data)
• "Compare youth unemployment in Brazil, Mexico and South Africa." (ilo_compare_countries)
• "How large is the gender pay gap, and where does the ILO publish it?" (ilo_search_indicators → ilo_get_data)
• "What share of employment in India is informal?" (ilo_search_indicators → ilo_get_data)
• "Which ILOSTAT dataflow has average monthly earnings by sex and economic activity?" (ilo_search_indicators)
• "Which country, age and sex codes can I filter this indicator by?" (ilo_list_dimension_values)
• "Give me a labour-market profile of Viet Nam." (ilo_country_labour_profile)

Ask in your words, not the ILO's. ILOSTAT is worded in British statistical English, and the catalogue is matched on the dataflow name — so the everyday or US word used to return nothing at all. Measured over the 1,212 dataflows of the official catalogue (2026-09-13), and fixed since 0.6.0: the search translates the term and tells you it did.

Comparison with the alternatives

Anyone who already works with ILOSTAT has good tools, and this server replaces none of them — it sits somewhere else in the chain: it answers the question at the point where the question is asked, inside the assistant, with source, vintage and licence attached to the answer. Detail, side-by-side examples and the measured numbers in docs/alternatives.md.

Do not use this server when you need a whole dataset rather than an answer (Rilostat's bulk download is the right tool), when the question is not labour statistics published by the ILO (education → UNESCO UIS, national accounts → IMF/World Bank), or when you need microdata: ILOSTAT publishes aggregates, and so does this server.

Sister servers, same design and same provenance block, for other official sources: IBGE (Brazilian statistics), BCB (Central Bank of Brazil), Senado (Brazilian Senate open data), SIH/SUS (Brazilian hospital admissions) and medical terminologies (ICD-11, ICD-10, LOINC, RxNorm, ATC, MeSH).

Use it (hosted — no setup)

Point any MCP client at the Streamable HTTP endpoint:

https://ilo.sidneybissoli.com/mcp

Claude Desktop / Claude Code and other clients with native remote support:

{  "mcpServers": {    "ilostat": {      "url": "https://ilo.sidneybissoli.com/mcp"    }  }}

For clients that launch MCP servers as a command, use the mcp-remote bridge:

{  "mcpServers": {    "ilostat": {      "command": "npx",      "args": ["-y", "mcp-remote", "https://ilo.sidneybissoli.com/mcp"]    }  }}

The ilo-mcp-server.sidneybissoli.workers.dev hostname is also served, as a secondary.

ChatGPT (Deep Research)

ChatGPT deep research (and company knowledge, and research workflows over the Responses API) only uses an MCP server that exposes exactly search and fetch — this server does, on top of the ilo_* tools. Point the connector at the hosted endpoint, no key required:

https://ilo.sidneybissoli.com/mcp

search ranks the query against the full ILOSTAT dataflow catalogue (~1,200 SDMX dataflows — employment, unemployment, wages, working time, informality, SDG labour indicators) and returns { id, title, url } (ind:<DATAFLOW_ID>, e.g. ind:DF_UNE_2EAP_SEX_AGE_RT); fetch returns the dataflow as readable Markdown — name, data vintage, dimensions and codelists, the ILO's default selection and how to query it with ilo_get_data — with the public ILOSTAT data explorer page as url, which is what ChatGPT cites. Both carry the same provenance block as every other tool, in structuredContent and _meta (the text channel is the contract's JSON). In ChatGPT's developer mode (Settings → Security and login → Developer mode) any tool is callable — the ilo_* tools remain the ones to use for data.

Run locally (stdio)

Prefer not to route queries through a third-party host? The same server also runs as a local stdio process that talks directly to the official ILOSTAT API — same 6 tools, resources and prompts, same limits, same provenance block, no Cloudflare in the loop.

No install needed — the package is on npm (ilo-mcp-server, Node ≥ 20):

{  "mcpServers": {    "ilostat": {      "command": "npx",      "args": ["-y", "ilo-mcp-server"]    }  }}

Or from source:

git clone https://github.com/SidneyBissoli/ilo-mcp-servercd ilo-mcp-servernpm installnpm run buildnode dist/cli.js   # serves MCP over stdio (Ctrl+C to stop)

(then point the client at node /path/to/ilo-mcp-server/dist/cli.js).

Differences from the hosted server, all due to the absence of Cloudflare bindings: the SDMX cache lives in process memory (structures and codelists are reused within a session, not across sessions); the search catalogue is downloaded from the official endpoint on the first search (its real retrieved_at is reported in provenance); no usage metrics, rate limit or auth. Logs go to stderr — stdout carries only the JSON-RPC stream. The repository Dockerfile builds this runtime (used by the Glama registry).

Tools

Typical flow: ilo_search_indicators → ilo_get_indicator_metadata / ilo_list_dimension_values to discover valid filter codes → ilo_get_data with country and period filters.

Every response carries the portfolio provenance block (@sbissoli/mcp-provenance, modes concise/detailed via the provenance_mode parameter) on three channels: structuredContent, namespaced _meta (com.sidneybissoli.ilostat/*) and a text footer. Since 1.1.0 the block includes retrieval, the origin diagnostic of the call — how many requests went to ILOSTAT, how many attempts they took, which anomalies were overcome and whether the answer is unstable — measured by the portfolio's common fetch (@sbissoli/mcp-upstream); null when the answer did not touch ILOSTAT (catalogue or cached structure).

Resources and prompts

Three resources (static, text/markdown, no upstream call) that a client can attach to the context before calling tools — they save the 2–3 discovery calls most sessions spend on "which dataflow, which codes":

Three prompts — ready-made workflows that chain the tools and end with the citation rules (arguments are strings; period arguments optional):

Every dataflow id quoted in the resources and prompts is checked against the catalogue seed by the test suite, so the documentation cannot point at an id the search would not find.

Behaviour and limits

• REF_AREA is required in ilo_get_data, up to 30 areas per call. The ILO gateway times out (HTTP 504) on unrestricted queries, so the server never issues one; for broad panels, split the areas into batches and/or paginate by period (start_period/end_period). The error message explains how.
• One live REST call per data query. Data is never cached — every ilo_get_data result is fetched from ILOSTAT at request time. Dataflow structures (TTL 24 h) and codelists (TTL 7 days, shared across dataflows) are cached.
• data_vintage is the dataflow's last-update date as published by the ILO (LAST_UPDATE annotation, normalised to ISO).
• retrieved_at is always the real instant of extraction from ILOSTAT, preserved alongside any cached value — never the build or response time. Cached responses say so (served_from_cache: true).
• The indicator catalogue is a local snapshot (~1,210 dataflows), refreshed periodically; its own retrieved_at is reported in the provenance of ilo_search_indicators, so its age is always visible.
• Every upstream call carries an identifiable User-Agent (service URL + contact), so ILO administrators can reach the operator.
• Every upstream call has a timeout and a retry policy (since 1.1.0, measured against the live API): 65 s per attempt — above the ~61 s the ILO gateway itself takes to reject an unrestricted query with 504, so a legitimate wide query is never cut short by this server —, up to 3 attempts on 5xx, 429 (honouring Retry-After) and network errors, 70 s in total per call. Timeouts, HTTP 504 (query too broad) and 404 (no observations) are never retried. What happened is reported in the provenance retrieval field.
• Language: English; timezone: UTC (ILO data is published in English).

Provenance fields

• derived — true only for real transformation (aggregation, server-computed rate, interpolation, harmonisation), always with a derivation_note; unit conversion and rounding do not count. This server does not transform values, so derived is always false (and absent from the concise block).
• notices (ilo_get_data) — each distinct value of OBS_STATUS (the SDMX status/disclaimer channel, e.g. "Break in series"), UNIT_MEASURE and UNIT_MULT (e.g. "Thousands": the unit and scale are part of the number) found in the rows, with the number of observations carrying it. Repeated in the text footer; absent when there is nothing to say. Technical per-observation attributes (DECIMALS etc.) stay on the rows (rows[].attributes).
• revision (since contract 1.3, server 1.6.4) — always current: ILOSTAT serves only its current release and may revise it later; on data answers the note says the ILO re-estimates modelled series (_2 ids) at each release, past years included.

Surface fingerprint, checkable from the registry

Every release publishes, in its MCP Registry entry, the sha256 of the surface this version serves (initialize, tools, resources, prompts) and which methods answer without a credential. A surface change without a version bump fails the build; the fingerprint in the registry lets a client check the same thing on its own side. Canonical form and procedure: SPEC.md of @sbissoli/mcp-surface. To check it yourself (Node 18+, no dependencies):

curl -sO https://raw.githubusercontent.com/SidneyBissoli/mcp-br-commons/main/packages/mcp-surface/exemplos/verify.mjsnode verify.mjs io.github.SidneyBissoli/ilo-mcp-server

Data license and attribution

• ILOSTAT data and metadata: CC BY 4.0 (since 2023-05-03; license verified 2026-08-04).
• ILO attribution in every response (citation field): International Labour Organization, ILOSTAT, https://ilostat.ilo.org/data/, accessed <date>.
• The ILO logo is not used. This service is not endorsed by the ILO.

Self-hosting / development

Everything below is only needed to run your own instance — it is not required to use the public server.

npm installnpm run typecheck && npm test   # offline suite (parsers, key, tools, output contract, resources/prompts, in-memory catalogue, vocabulary, eval fixtures)npm run dev                     # http://localhost:8787/mcp (Worker)npm run build && npm start      # stdio runtime (dist/cli.js)
# Catalogue seed (D1) — required before first use:node scripts/seed-catalog.mjs   # downloads via curl and generates scripts/seed-catalog.sqlnpx wrangler d1 execute ilostat-catalog --local  --file=scripts/seed-catalog.sqlnpx wrangler d1 execute ilostat-catalog --remote --file=scripts/seed-catalog.sql
npm run deploynode scripts/smoke-mcp.mjs      # smoke test against production (initialize → 6 tools → search → fetch → errors)npm run manifest:lhm            # regenerate tools/resources/prompts in lhm.plugin.json from the real server# (the seed also writes tests/fixtures/catalog-ids.txt — the versioned id list the tests check resources/prompts against)

Bindings (see wrangler.jsonc): KV SDMX_CACHE, D1 CATALOG_DB, Durable Object USAGE (SQLite-backed usage counters), CF_VERSION_METADATA. Optional Bearer auth (wrangler secret put API_KEY); token-bucket rate limit per IP.

Notes for operators:

• ILOSTAT returns JSON only when negotiated via the Accept header (application/vnd.sdmx.{structure,data}+json); ?format= is ignored and returns XML.
• The ILO gateway answers HTTP 500 (languageTag1) to the Accept-Language: * header that Node's fetch (undici) sends by default; every upstream call therefore sets Accept-Language: en explicitly (Cloudflare's runtime sends no such header, so the Worker was never affected). It also expects an identifiable User-Agent.
• Catalogue refresh is manual (no cron): quarterly, or immediately if a dataflow that exists upstream does not show up in search. Procedure: the three seed commands above. Data queries are always live, so only the search catalogue can age — and its age is exposed in provenance.

Evals

@sbissoli/mcp-evals: 24 fixtures in evals/fixtures/queries.ts, validated offline in npm test. The run with a real model (npm run eval) uses the Anthropic API and needs ANTHROPIC_API_KEY (without it, it exits with instructions). Run of 2026-08-07: top-1 100% (24/24) — evals/results/.

End-to-end: 10 complex questions with a single verifiable answer in evals/e2e/evaluation.xml, answers validated manually against production (evals/e2e/validacao-respostas.md). Run of 2026-08-07 (Sonnet): 9/10 exact string; 10/10 substantive — evals/results/2026-08-07-e2e.md.

Endpoints

Security

Snyk Agent Scan (2026-08-07): passed — report in security/.

License

Code: MIT. Data: ILOSTAT, CC BY 4.0 (see "Data license and attribution" above).

Privacy

Privacy policy of the hosted service: PRIVACY.md.

Contact

Sidney da Silva Pereira Bissoli — [email protected]. This service is not endorsed by the ILO.