Data Reunion

What the Data Reunion MCP server does

The Data Reunion MCP server gives MCP-compatible AI clients access to datasets published through Réunion Open Data at data.regionreunion.com. Its main purpose is structured dataset discovery and querying rather than general web browsing. The available data covers areas such as regional services, transport, environment, and the economy, subject to what the Réunion Open Data catalog currently publishes.

The server exposes three pack-specific tools. search_datasets finds relevant datasets by keyword and returns identifiers, titles, themes, and record counts. dataset_info describes one dataset, including its fields, schema, themes, and number of records. query retrieves or analyzes records with ODSQL options for filtering, grouping, selecting, ordering, and pagination.

How it works

A typical workflow starts with search_datasets. Use a returned dataset_id with dataset_info before writing a query so the agent can learn the available column names and data structure. The agent can then call query with a where filter, group_by and select expressions for aggregation, order_by for sorting, and limit or offset for pagination.

The remote MCP endpoint is available at https://gateway.pipeworx.io/data-reunion/mcp. A connection to this scoped endpoint lists the three Data Reunion tools along with shared Pipeworx gateway tools. Those shared tools include routing and discovery functions such as ask_pipeworx, so the endpoint can also direct broader questions across the Pipeworx catalog. The full gateway endpoint is available at https://gateway.pipeworx.io/mcp when direct access to all catalog packs is preferred.

The standalone package provides the same source-specific tools over MCP stdio and does not add the gateway’s shared meta-tools. It is useful when the client should connect directly to this data source without a gateway round trip.

Setup and configuration

For a remote connection, add the gateway URL to an MCP client configuration:

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

The README names Claude Desktop, Cursor, and Windsurf as supported client examples.

For local execution, configure the package as a stdio server:

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

You can also verify that the local process starts with npx -y @pipeworx/mcp-data-reunion. The package communicates through standard input and output and requires no Pipeworx account.

Tools and capabilities

• search_datasets: Locate catalog datasets by keyword and return IDs, names, themes, and record counts.
• dataset_info: Inspect a dataset’s fields, schema, themes, and record count.
• query: Apply ODSQL filtering, selection, grouping, ordering, and pagination to dataset records.

Without an MCP client, the gateway also exposes HTTP tool endpoints. For example, a search can be submitted to /v1/tools/data_reunion_search_datasets with a JSON body containing a query value. Tool details can be inspected through the corresponding gateway tool metadata endpoint.

Limitations and notes

The pack-specific tool list does not represent every tool returned by the remote scoped endpoint. Gateway connections add shared Pipeworx tools, and the exact scope is reported by the MCP initialize response. The standalone stdio mode exposes only the three Data Reunion tools and does not provide ask_pipeworx routing.

Queries depend on the selected dataset’s actual schema. Because field names vary, dataset_info should normally precede query.

The Data Reunion MCP server is focused on the Réunion Open Data catalog. Questions requiring sources outside that catalog may need Pipeworx routing or another MCP server.