Cms

What Cms does

Cms exposes selected CMS Open Data operations through the Model Context Protocol. Agents can discover datasets by searching titles, descriptions, and keywords in the CMS DCAT catalog, then use the returned dataset UUID with the metadata and data-retrieval tools.

The available catalog covers areas such as Medicare and Medicaid providers, spending, enrollment, drug pricing, quality measures, hospitals, nursing homes, and accountable care organizations. Cms is an independent integration and is not affiliated with, endorsed by, or published by the Centers for Medicare & Medicaid Services.

How it works

The hosted MCP endpoint is https://gateway.pipeworx.io/cms/mcp. An MCP client connects to that URL and can call the CMS-specific tools. A separate full-gateway URL, https://gateway.pipeworx.io/mcp, exposes the broader Pipeworx catalog directly.

The normal workflow is:

1. Call search_datasets with a keyword to find matching dataset IDs and titles.
2. Pass a selected UUID to dataset_info to review its description, row count, modification date, themes, keywords, and downloadable resources.
3. Call get_dataset with that UUID to retrieve rows.

The gateway also supplies shared Pipeworx meta-tools on hosted pack connections. These can route broader questions through ask_pipeworx, while the local npm bridge lists only the CMS pack by default unless meta-tools are enabled.

Setup and configuration

For a hosted connection, add the CMS endpoint to an MCP client configuration. Claude Desktop, Cursor, and Windsurf are shown as supported examples:

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

The package @pipeworx/mcp-cms provides a local stdio bridge. Start it with npx so calls are forwarded to the same hosted gateway:

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

The bridge requires network access. It can make anonymous calls, or use PIPEWORX_API_KEY for an account-backed quota. Set PIPEWORX_META_TOOLS=1 when the local process should also list the gateway-wide meta-tools. For HTTPS proxy environments, Node may require NODE_USE_ENV_PROXY=1 in versions that support that option.

The tools are also callable over HTTP. For example, POST https://gateway.pipeworx.io/v1/tools/cms_search_datasets accepts a JSON body such as {"keyword":"hospital"}.

Tools and capabilities

• search_datasets searches CMS dataset titles, descriptions, and keywords and returns IDs for subsequent calls.
• dataset_info returns dataset metadata and links or details for downloadable resources such as CSV files and data dictionaries.
• get_dataset returns row objects keyed by the dataset's columns.
• Retrieve data in pages with size and offset.
• Search within a dataset using full-text keywords.
• Apply exact-match column filters with a filters object, such as {"State":"TX"}.

Limitations and notes

Anonymous use of the npm bridge is capped at 50 calls per day. A free account raises that limit to 200 calls per day; authenticated calls use a Bearer token and are metered according to the gateway account.

The local bridge is not an offline implementation: it forwards requests to the Pipeworx gateway and therefore needs network access. Pack endpoints include shared gateway meta-tools in addition to their pack-specific tools, while the local bridge hides those tools unless configured otherwise. Dataset identifiers must be obtained from the CMS catalog or otherwise supplied as CMS UUIDs before retrieving metadata or rows.