Data Paris

What the Data Paris MCP server does

The Data Paris MCP server connects AI agents to Paris Open Data at opendata.paris.fr through the OpenDataSoft platform. It is intended for finding and analyzing public datasets rather than for writing data back to the source.

The available tools cover three stages of a typical workflow:

• search_datasets finds datasets using keywords such as mobility, trees, events, city services, urban planning, or environmental topics. Results include dataset IDs, titles, themes, and record counts.
• dataset_info returns metadata for a selected dataset, including its fields, schema, themes, and record count.
• query retrieves or summarizes records from a dataset using ODSQL. It supports conditions, grouping and selected fields, ordering, and offset-based pagination.

Use dataset_info before writing a query when the field names are unknown. The returned schema helps determine which columns can be used in filters, grouping, or selected output.

How it works

The hosted option uses the Pipeworx gateway. Add the streamable HTTP endpoint https://gateway.pipeworx.io/data-paris/mcp to an MCP client. The gateway exposes the pack's three tools and also includes shared Pipeworx meta-tools, so the tool list is larger than the Data Paris-specific list. Its ask_pipeworx tool can route questions across the wider Pipeworx catalog.

The full gateway endpoint, https://gateway.pipeworx.io/mcp, is another connection option. Both hosted URLs reach the same gateway and data sources; they differ in which pack tools are listed directly.

The Data Paris MCP server can also run locally as a stdio process. In that mode, it exposes only the pack's own tools and does not use the gateway's shared meta-tools or routing. The local package and hosted gateway access the same source data.

Setup and configuration

For a hosted connection, add the following MCP server configuration:

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

For local stdio use, configure npx to download and run the package:

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

Running npx -y @pipeworx/mcp-data-paris directly starts the local MCP process.

Without an MCP client, individual tools can be called over HTTP. For example, the hosted REST-style endpoint for dataset search is POST https://gateway.pipeworx.io/v1/tools/data_paris_search_datasets with a JSON body containing a query value.

Tools and capabilities

A practical sequence is to search for a subject, inspect the chosen dataset's metadata, then query its records. ODSQL queries can narrow records with where, produce grouped or selected results with group_by and select, order results with order_by, and divide large responses with limit and offset.

The server fits questions that depend on Paris-specific public datasets, such as identifying relevant mobility data, examining tree records, or summarizing an urban-planning dataset. It does not replace schema inspection: query field names depend on the selected dataset.

Limitations and notes

The hosted pack endpoint includes shared gateway tools in addition to the three pack tools. The exact combined tool scope can change, so the connection's initialize response is authoritative. Choose the local stdio package when only the Data Paris tools are wanted, or use the full gateway when direct access to all catalog packs is preferable.