Opendatasoft

What Opendatasoft MCP server does

The Opendatasoft MCP server gives an MCP client access to public datasets published on Opendatasoft portals. It is designed for portals that expose the Opendatasoft Explore API, including the default public.opendatasoft.com portal and other instances such as data.paris.fr or data.economie.gouv.fr.

The server supports discovery and retrieval workflows. An agent can search a portal catalog, inspect a dataset’s metadata, query rows from a dataset, enumerate distinct values for a facet, or inspect metadata for the portal itself.

How it works

The Opendatasoft MCP server uses the Opendatasoft Explore API at paths under https://<instance>/api/explore/v2.1/catalog/, including dataset record endpoints. Each tool accepts an optional instance value. When it is omitted, requests go to public.opendatasoft.com; supplying a hostname directs the request to a different Opendatasoft portal.

Two connection modes are available. The hosted endpoint is https://gateway.pipeworx.io/opendatasoft/mcp. It exposes the Opendatasoft tools along with shared Pipeworx meta-tools, including ask_pipeworx, discovery, search, and memory tools. The full gateway endpoint is also available when an agent needs direct access to every source. Alternatively, the package can run as a local stdio MCP server and exposes only the Opendatasoft tool set.

The gateway can also be called without an MCP client through HTTP. Tool calls use endpoints such as /v1/tools/opendatasoft_datasets, with JSON request bodies.

Setup and configuration

For a hosted connection, add the following MCP server entry to a compatible client:

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

For local stdio operation, use the npm package:

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

The local process communicates over stdin and stdout and does not need a Pipeworx account. The hosted route and the local package use the same source tools, but only the hosted route includes the gateway’s shared meta-tools. No environment variables are documented as required.

Tools and capabilities

The Opendatasoft MCP server provides these operations:

• datasets: search an instance catalog using text, row count, offset, sorting, and facets.
• dataset: retrieve metadata for one dataset by identifier.
• records: query dataset rows with filters, selected fields, grouping, ordering, limits, and offsets.
• facets: retrieve distinct values for a dataset facet.
• instance_info: inspect metadata for an Opendatasoft instance.

For natural-language requests on the gateway, ask_pipeworx can select a suitable tool and populate its arguments. Direct tool use is preferable when the agent needs explicit control over dataset IDs, filters, pagination, or the target instance.

Limitations and notes

The documented scope is public Opendatasoft data. The default target is a public mash-up portal, but results depend on the selected instance and its available datasets. The material does not document authentication for private portals, write operations, uploads, or dataset administration.

A pack-specific hosted connection lists more tools than the five Opendatasoft operations because shared gateway tools are included. The exact tool list can change; the connection’s initialize response is the authoritative scope for that connection. The local stdio mode avoids those shared tools and does not provide ask_pipeworx routing.

The package is released under the MIT license.