Overview

EU statistics from the Eurostat catalogue: economy, demography, trade, health, and NUTS regional data. Search and browse the catalogue by keyword or theme, inspect dataset dimensions, then query a slice or bulk-download a whole dataset. Runs as a stdio process, a local Streamable HTTP server, or the public hosted endpoint above.

The catalogue spans two Eurostat hosts. Most datasets come from the dissemination API; the DS-* collections — detailed trade by CN8, HS, SITC, BEC and CPA (Comext), and PRODCOM industrial production — come from the Comext dissemination host. Every tool routes a DS-* code there by its prefix, in any case, so the same calls work on both.

Tools

Two of the eight are listed only when the dataframe canvas is enabled (CANVAS_PROVIDER_TYPE=duckdb).

Resources

The same metadata is available through eurostat_get_dataset_info for tool-only clients.

Capability reference

eurostat_search_datasets _tool

• query tokens are ANDed case-insensitively across each dataset's label, theme breadcrumb, and code; limit 1–100 (default 20), paged by passing nextCursor back as cursor
• Results carry code, label, type (dataset / table), period coverage, obsCount, lastUpdated, and themePath; totalMatches counts every match
• Searches the dissemination table of contents plus the Comext host's dataflow list, whose DS-* entries carry lastUpdated but no period coverage or obsCount. A collection on neither, such as the legacy PRODCOM DS-056120, is not disseminated and cannot be reached
• Fails as no_match — for a query naming such a DS-* code, saying the collection is not disseminated rather than suggesting a broader search — or invalid_cursor when a cursor is reused with a different query or after the catalogue refreshes

eurostat_browse_themes _tool

• Omit theme_code for the top-level theme folders; pass a folder code for its immediate children
• The Comext collections sit in two folders the table of contents lacks: ext_go_detail (detailed trade) under ext_go, and prom (PRODCOM) under icts
• Items carry code, label, type (folder / dataset / table), and hasChildren, with a parentPath breadcrumb; otherPlacements names other branches that file the same folder code
• Fails as not_found for an unknown code, not_a_folder for a dataset or table code

eurostat_get_dataset_info _tool

• One dataset_code; returns every dimension with valuesCount and up to 10 sampleValues, plus timeRange, obsCount, lastUpdated, and the ESMS metadataUrl
• The Comext host reports no timeRange or obsCount for a DS-* collection, so both read as unreported. Its structure carries a label for every code, 23 MB for DS-045409: parsed metadata is cached per dataset for an hour, within a 64 MiB memory budget, so only the first call pays for it
• Fails as not_found for an unknown dataset, or upstream_fault when Eurostat returns a structure or content constraint the server cannot read; a failed read is not cached, so a later call downloads it afresh

eurostat_get_dimension_values _tool

• dataset_code plus dimension; returns the dataset-available code/label pairs in values, with totalCount
• At most 2,000 values come back inline, in Eurostat's order. A longer list — the 37,069 CN8 codes of DS-045409, a daily time dimension, the largest airport-pair lists — is cut there, with truncated, shown, cap, and a notice saying where the rest is
• For geo, geo_level picks a NUTS level (aggregate, country by default, nuts1, nuts2, nuts3)
• canvas_id also stages every value, past the inline cap too, on that canvas as a code / label table (canvasId / tableName / stagedRowCount), to search with SQL or to label a download's code-only columns through a join; this tool never starts a canvas of its own, and ignores canvas_id on a deployment without one
• Fails as not_found for an unknown dataset or dimension, no_results for a NUTS level with no values, conflicting_params for geo_level on any other dimension, upstream_fault for a structure or content constraint the server cannot read, or canvas_not_found for an unknown or expired canvas_id

eurostat_query_dataset _tool

• dataset_code plus filters ({dimension_code: [values]}, codes in any case), geo_level, and either since_period / until_period (YYYY, YYYY-MM, YYYY-MM-DD, YYYY-Qn, YYYY-Sn, YYYY-Tn, YYYY-Mnn, YYYY-Wnn, YYYY-Dnnn; extra leading zeros after the letter are dropped, so 2020-Q01 is sent as 2020-Q1, and YYYY-A1 is sent as YYYY) or last_n_periods, which counts back from the dataset's latest period rather than the slice's; preview_limit 1–500 (default 50) sets the inline row count; lang is EN, FR, or DE
• Each observation carries code/label pairs per dimension, a nullable value, an OBS_FLAG status, and a CONF_STATUS confStatus; obsCount, missingObsCount, and timeRange cover the whole match, truncated is true above 5,000 observations, and unmatchedValues names filter values that matched nothing
• PRODCOM publishes its flag and unit indicators as text in the value itself: :C comes back as a null value with confStatus C, and anything else, such as the unit KG, verbatim in valueText with a null value
• Fails as not_found, no_results (naming the filter values that matched nothing and, up to the newest 24 with a full count, the selected periods that carry no value), invalid_dimension, invalid_period (a malformed or non-existent period, or a since_period that starts after until_period ends, rejected before any request), conflicting_params (geo filter with geo_level, or a period range with last_n_periods), a non-retryable async_response for a query too large to serve inline, or canvas_not_found for an unknown or expired canvas_id

eurostat_download_dataset _tool

• dataset_code plus the same filters map and since_period / until_period in the same period forms (no last_n_periods: the TSV keeps a column per period unless a range drops it); preview_limit 1–500 (default 50)
• Returns rowCount (echoed as totalCount), missingCount, periodRange, and bytesRead for the whole download; the EUROSTAT_BULK_MAX_BYTES budget (default 50 MiB) stops a transfer mid-stream and returns what arrived with budgetExceeded: true
• On a DS-* table, obs_value_text holds a value published as text (PRODCOM's KG), and a PRODCOM :C lands in conf_status
• Fails as not_found, invalid_dimension (including a range wholly outside the dataset's coverage), invalid_period (a malformed or non-existent period, or a since_period that starts after until_period ends, rejected before any request), filter_arity, a non-retryable async_queued when Eurostat queues the extraction, a non-retryable extraction_too_big when Eurostat refuses it as too large (past its 5,000,000-row limit, or an unfiltered download of a large Comext collection) with the dimensions left to filter, no_results (including a range inside the dataset's coverage that misses the filtered series), upstream_fault, or canvas_not_found for an unknown or expired canvas_id

eurostat_dataframe_describe _tool

• canvas_id from a staging response; lists each table's name, rowCount, and typed columns, plus canvas and table expiresAt (sliding CANVAS_TTL_MS, default 24h); fails as canvas_not_found for an unknown or expired ID, or canvas_disabled on a deployment without a canvas
• eurostat_query_dataset tables carry a code and a _label column per dimension; eurostat_download_dataset tables carry codes only, plus time. Both share obs_value, obs_flag, obs_flag_label, conf_status, and conf_status_label, so they join on dimension codes and time; a DS-* table from either adds obs_value_text
• eurostat_get_dimension_values tables carry code and label: join one to a download on its dimension column (d.geo = g.code) to label it

eurostat_dataframe_query _tool

• One read-only SELECT per call; chained statements, other verbs, and functions that read files or external data are rejected
• Returns columns, rows, rowCount, and truncated, which is true past CANVAS_DEFAULT_ROW_LIMIT (default 10,000) rows; 64-bit integers, COUNT(*) included, arrive as strings. Fails as missing_table, canvas_not_found, or canvas_disabled

eurostat://dataset/{dataset_code} _resource

• Same payload as eurostat_get_dataset_info, returned as application/json
• dataset_code comes from eurostat_search_datasets or eurostat_browse_themes

Features

Built on @cyanheads/mcp-ts-core: stdio and Streamable HTTP transports, pluggable auth (none / jwt / oauth), swappable storage (in-memory, filesystem, Supabase, Cloudflare KV/R2/D1), structured logging with optional OpenTelemetry tracing.

Eurostat-specific:

• The Statistics API (JSON-stat 2.0) for slices, the SDMX 2.1 TSV bulk endpoint for whole datasets at roughly half the bytes, and the catalogue TOC cached in memory for 12 hours
• The Comext host's DS-* collections through the same tools: detailed trade by CN8, HS, SITC, BEC and CPA, and PRODCOM. The trade flows mix annual and monthly series in one dataset, so filter freq; product, reporter and partner carry aggregates (TOTAL, EU27_2020, EXT_EU27_2020) that double-count when summed with their members. The Comext host serves no large collection unfiltered
• NUTS geo-level filtering (aggregate / country / nuts1 / nuts2 / nuts3) on eurostat_query_dataset and eurostat_get_dimension_values
• OBS_FLAG (provisional, estimated, and so on) and CONF_STATUS (confidentiality) decoded into separate fields, with the same codes from JSON-stat and the bulk TSV
• Eurostat's too-large-to-serve responses (HTTP-200 async warnings, HTTP 413, SOAP fault 413, SOAP queue tickets) come back as non-retryable async_response, extraction_too_big, or async_queued errors that name the dataset's own dimensions left to filter, not timeouts
• Inline rows are capped by preview_limit (at most 500). With CANVAS_PROVIDER_TYPE=duckdb, a query match above 5,000 observations, or any bulk download, is staged whole as a SQL table (canvasId / tableName / stagedRowCount) for the dataframe tools, and canvas_id stages the next result beside earlier ones for joins — including a dimension's code/label list from eurostat_get_dimension_values, which labels a bulk table's codes. An empty download fails as no_results without creating or touching a canvas. The .mcpb bundle strips the native DuckDB binding, so SQL analytics need the npm, Docker, or from-source install

Agent-friendly output:

• Typed error contracts: every declared failure carries a reason and a recovery.hint naming the next tool to call
• Next-step hints: eurostat_search_datasets and eurostat_browse_themes return a nextStep pointing at the follow-up call
• Unknown stays unknown: counts and period bounds Eurostat doesn't report (obsCount, timeRange.start / end, lastUpdated) are omitted rather than zeroed
• Effective-query echo: appliedFilters on eurostat_query_dataset and appliedQuery (with the SDMX url) on eurostat_download_dataset, plus a notice when the inline preview omits rows, naming the staged table when there is one; eurostat_query_dataset's notice also names any filter value that matched nothing

Getting started

Public Hosted Instance

A public instance is available at https://eurostat.caseyjhand.com/mcp — no installation required. Point any MCP client at it via Streamable HTTP:

{
  "mcpServers": {
    "eurostat-mcp-server": {
      "type": "streamable-http",
      "url": "https://eurostat.caseyjhand.com/mcp"
    }
  }
}

Self-Hosted / Local

Add the following to your MCP client configuration file.

{
  "mcpServers": {
    "eurostat-mcp-server": {
      "type": "stdio",
      "command": "bunx",
      "args": ["@cyanheads/eurostat-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info"
      }
    }
  }
}

Or with npx (no Bun required):

{
  "mcpServers": {
    "eurostat-mcp-server": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@cyanheads/eurostat-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info"
      }
    }
  }
}

Or with Docker:

{
  "mcpServers": {
    "eurostat-mcp-server": {
      "type": "stdio",
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "MCP_TRANSPORT_TYPE=stdio", "ghcr.io/cyanheads/eurostat-mcp-server:latest"]
    }
  }
}

For Streamable HTTP, set the transport and start the server:

MCP_TRANSPORT_TYPE=http MCP_HTTP_PORT=3010 bun run start:http
# Server listens at http://localhost:3010/mcp

Prerequisites

• Bun v1.4.0 or higher (or Node.js v24+).
• No API key: Eurostat's dissemination API and its Comext host are public.

Installation

1. Clone the repository:

git clone https://github.com/cyanheads/eurostat-mcp-server.git

2. Navigate into the directory:

cd eurostat-mcp-server

3. Install dependencies:

bun install

4. Configure environment (optional, every setting has a default):

cp .env.example .env

Configuration

See .env.example for the full list of optional overrides.

Running the server

Local development

• Build and run the production version:

  sh

  # One-time build
  bun run rebuild
  
  # Run the built server
  bun run start:http
  # or
  bun run start:stdio
  

• Run checks and tests:

  sh

  bun run devcheck  # Lints, formats, type-checks, and more
  bun run test      # Runs the test suite
  bun run lint:mcp  # Validates MCP definitions against spec
  

Docker

docker build -t eurostat-mcp-server .
docker run --rm -p 3010:3010 eurostat-mcp-server

The Dockerfile defaults to HTTP transport, stateless session mode, and logs to /var/log/eurostat-mcp-server. OpenTelemetry peer dependencies are installed by default — build with --build-arg OTEL_ENABLED=false to omit them.

Project structure

Development guide

See CLAUDE.md for development guidelines and architectural rules. The short version:

• Handlers throw, framework catches — no try/catch in tool logic
• Use ctx.log for logging, ctx.state for storage
• Register new tools and resources in the createApp() arrays in src/index.ts
• Wrap external API calls: validate raw → normalize to domain type → return output schema; never fabricate missing fields

Contributing

Issues are welcome. Run checks and tests before submitting:

bun run devcheck
bun run test

License

This project is licensed under the Apache 2.0 License. See the LICENSE file for details.