OECD MCP Server

Public Hosted Server: https://oecd.caseyjhand.com/mcp

Tools

Five discovery and data tools plus two SQL analytics tools for large query results:

oecd_list_agencies

Entry point for discovery — enumerate OECD's statistical departments before searching.

• Returns agency IDs (e.g. OECD.SDD.NAD, OECD.ELS.SPD, OECD.EDU.IMEP) and dataflow counts

• Each agency carries the name of its directorate — OECD.CTP.TPS is the Centre for Tax Policy and Administration, OECD.SDD.NAD the Statistics and Data Directorate — so a department can be picked without decoding the identifier

• Publishers outside OECD that ship dataflows through the same catalog (ESTAT, IAEG-SDGs) carry no directorate

• Useful for scoping oecd_search_datasets by department (national accounts, labour, education, etc.)

oecd_search_datasets

Search the full catalog of 1,500+ OECD dataflows by keyword or department.

• Token-matching across dataflow names and descriptions — reaches datasets whose name never carries the term, so inflation returns Economic Outlook 119 and poverty returns Income inequality - Regions

• Each result reports matched_in (name, description, or both) and a plain-text description trimmed to 240 characters

• Optional agency_id filter scopes results to a specific statistical department

• limit (1–100) and offset page through the match list; total_matches reports the full count

• Returns flow_ref values (e.g. OECD.SDD.NAD,DSD_NAAG@DF_NAAG_I) — pass directly to oecd_get_dataset_info or oecd_query_dataset. A handful of dataflows are catalogued without a datastructure prefix and come back in the bare {agencyID},{df_id} form (OECD.TAD.ARP,DF_AEI2024_DASHBOARD); both forms are accepted everywhere a flow_ref is

• Fetches and filters in-memory; the full catalog is ~5.9 MB and bounded (OECD adds datasets weekly, not continuously)

oecd_get_dataset_info

Inspect a dataflow's structure before querying.

• Returns all dimensions in key order (position 1, 2, 3 …) — dimension order is required to construct the dot-delimited key for oecd_query_dataset

• Each dimension carries its concept name from the datastructure's concept scheme, so INSTR_ASSET reads as "Financial instruments and non-financial assets" rather than repeating the id. A dimension the scheme does not cover keeps the id

• Shows codelist references for each dimension — pass to oecd_get_dimension_values to resolve human-readable names to SDMX codes

• Surfaces NonProductionDataflow flag — marks experimental or deprecated dataflows

• Resolves a flow_ref whose id prefix names no datastructure of its own by asking the dataflow for its structure — OECD.CFE.EDS,DSD_REG_LAB@DF_RATES is backed by DSD_REG_LABOUR, and answers here rather than reporting the dataflow as missing

• Required before calling oecd_query_dataset on an unfamiliar dataflow

oecd_get_dimension_values

Resolve human-readable names (countries, measures) to SDMX codes.

• Returns code + label pairs for a single dimension (e.g. REF_AREA → USA/United States, DEU/Germany)

• query matches a case-insensitive substring against both the code and its label, so PA and percent each reach PA / Percent per annum

• limit (1–500, default 50) and offset page the matching list. Both client surfaces carry the same page, so a 1,164-code dimension like UNIT_MEASURE no longer ships 66 KB of pairs to structuredContent to find one code

• When matches remain beyond the page, the response reports the full match count and how to reach the rest

oecd_query_dataset

Fetch observations from an OECD dataflow filtered by dimension key and time range.

• Accepts a dot-delimited key (e.g. A.USA+DEU.B1GQ_R.PC.) where empty segments are wildcards and + separates multiple values

• Optional start_period / end_period bound the time range (ISO format: 2010, 2010-Q1)

• Decodes SDMX-JSON index notation (0:0:2:3:0) into human-readable row objects with dimension labels

• Observation attributes (UNIT_MULT, OBS_STATUS, PRICE_BASE, DECIMALS, …) each become their own column, so an estimated or break-flagged point is distinguishable from a confirmed one

• value arrives already multiplied by the observation's UNIT_MULT — a GDP figure OECD publishes as 26054.614 billions comes back as 26054614000000. Every row carries value_scale, the power of ten applied; divide by it for the figure as OECD published it

• Every response row includes source: "OECD" per OECD terms of use

• Small results (few countries, narrow time range): every observation is returned inline, in structuredContent and in the rendered table alike — no canvas_id, and truncated is omitted rather than set to false

• Large results (multi-country, multi-year time-series) with CANVAS_PROVIDER_TYPE=duckdb: a leading preview slice plus canvas_id + truncated: true — use oecd_dataframe_describe to list tables, then oecd_dataframe_query for SQL analytics

• Large results without DataCanvas: there is nowhere to stage the remainder, so every observation still comes back in structuredContent, while the rendered table stops at the same preview budget a canvas would have used — the response reports content_table_capped and the number of rows it showed. Narrow the key or the start_period / end_period range to shrink the result itself

oecd_dataframe_describe / oecd_dataframe_query

SQL analytics over observation data staged by oecd_query_dataset.

When oecd_query_dataset returns truncated: true, the full result is staged on a DuckDB-backed DataCanvas. Pass the canvas_id to:

• oecd_dataframe_describe — list staged table names and their columns. Run this first to discover the schema before writing SQL.

• oecd_dataframe_query — run a single-statement SQL SELECT. Supports aggregates, window functions, GROUP BY, ORDER BY, and standard DuckDB SQL.

Requires CANVAS_PROVIDER_TYPE=duckdb. Read-only: writes, DDL, and system catalog access are rejected.

Typical workflow for a large query:

oecd_query_dataset → { canvas_id, table_name, truncated: true, rows: [preview...] }
  → oecd_dataframe_describe(canvas_id) → table/column names
  → oecd_dataframe_query(canvas_id, "SELECT REF_AREA, AVG(value) FROM spilled_... GROUP BY REF_AREA")

Related MCP server: global-education-mcp

Resources

{flow_id} is the combined {dsd_id}@{df_id} string with @ percent-encoded as %40, or the bare {df_id} for a dataflow catalogued without a datastructure prefix. Example: oecd://dataflow/OECD.SDD.NAD/DSD_NAAG%40DF_NAAG_I.

All resource data is also reachable via tools. Use oecd_get_dataset_info for the same content.

Features

Built on @cyanheads/mcp-ts-core:

• Declarative tool, resource, and prompt definitions — single file per primitive, framework handles registration and validation

• Unified error handling — handlers throw, framework catches, classifies, and formats

• Pluggable auth: none, jwt, oauth

• Swappable storage backends: in-memory, filesystem, Supabase, Cloudflare KV/R2/D1

• Structured logging with optional OpenTelemetry tracing

• STDIO and Streamable HTTP transports

OECD-specific:

• Keyless access — no API key required; OECD SDMX 2.1 REST API is fully public

• Covers 1,500+ dataflows across 20+ OECD statistical departments (national accounts, employment, inflation, trade, education, health, environment, taxation, inequality)

• Delegated dataflows resolved end to end — the entries OECD catalogues on one service root but defines on another (Trade in Value Added, the DAC creditor-reporting aid series) follow the catalog's own link for structure, codes, and observations, with the target checked against the configured origin before any request goes out

• Codes read at the revision the dataflow references — a codelist moves on independently of the datastructures using it, so a dimension's values come from the version its structure names rather than the endpoint's current latest, and never include a code the dimension rejects

• AllDimensions observation mode — one-pass SDMX-JSON decoding into flat row objects; no nested series key reconstruction

• oecd_query_dataset materializes large observation sets (multi-country time-series) on a DuckDB DataCanvas for in-conversation SQL analytics

• OECD source attribution (source: "OECD") on every observation row per OECD terms of use

Agent-friendly output:

• Workflow-aware tool surface — flow_ref from search flows directly into info, values, and query tools without reconstruction

• Spill signaling — truncated: true + canvas_id tells the agent to switch to SQL instead of parsing a truncated inline list

• Full SDMX decoding server-side — agents see { REF_AREA: "United States", MEASURE: "Gross domestic product", UNIT_MULT: "Billions", value: 26054614000000, value_scale: 1000000000 }, not raw index arrays

Getting started

Public Hosted Instance

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

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

Self-Hosted / Local

Add the following to your MCP client configuration file.

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

Or with npx (no Bun required):

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

Or with Docker:

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

To enable DataCanvas SQL analytics for large query results, add CANVAS_PROVIDER_TYPE=duckdb:

{
  "mcpServers": {
    "oecd-mcp-server": {
      "type": "stdio",
      "command": "bunx",
      "args": ["@cyanheads/oecd-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "CANVAS_PROVIDER_TYPE": "duckdb"
      }
    }
  }
}

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 v1.3.14 or higher (or Node.js v24+).

• No API key required — OECD SDMX is a free, public API.

Installation

1. Clone the repository:

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

2. Navigate into the directory:

cd oecd-mcp-server

3. Install dependencies:

bun install

4. Configure environment:

cp .env.example .env
# edit .env — most vars are optional; no API key required

Configuration

All configuration is validated at startup via Zod schemas in src/config/server-config.ts. Key environment variables:

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

Running the server

Local development

• Build and run:

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

• Run checks and tests:

  bun run devcheck   # Lint, format, typecheck, security
  bun run test       # Vitest test suite
  bun run lint:mcp   # Validate MCP definitions against spec

Docker

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

The Dockerfile defaults to HTTP transport, stateless session mode, and logs to /var/log/oecd-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 request-scoped logging, ctx.state for tenant-scoped storage

• Register new tools and resources via the barrels in src/mcp-server/*/index.ts

• Wrap external API calls: validate raw SDMX-JSON → normalize to domain type → return output schema; never fabricate missing fields

Contributing

Issues and pull requests are welcome. Run checks and tests before submitting:

bun run devcheck
bun run test

License

Apache-2.0 — see LICENSE for details.