NDIS Provider Feed

What NDIS Provider Feed MCP server does

NDIS Provider Feed MCP server makes data from the public NDIS Provider Register available to MCP-compatible agents. It restructures the source bulk CSV from one row per outlet into one record per provider, keyed by ABN, with outlet details nested under that provider. The feed includes business and legal names where available, entity type, registration status, expiry date, service categories, registration-condition flags, website information, head-office details, outlets, verification timestamps, and source metadata.

The source data is published by the NDIS Quality and Safeguards Commission under CC BY 4.0 AU. The project itself is not affiliated with or endorsed by the Commission or the NDIA. It is refreshed weekly through an automated process with validation checks before a new feed is published.

How it works

The hosted service uses a streamable HTTP MCP endpoint:

https://ndis-provider-feed-125475360011.australia-southeast1.run.app/mcp

The endpoint has no authentication. It is an MCP endpoint rather than a browser page, so opening the URL directly can return a Bad Request response. MCP clients should connect using streamable HTTP. Claude Code can add it with claude mcp add --transport http, while Claude Desktop or claude.ai can use the endpoint as a custom connector.

Searches return currently Approved providers by default and sort results alphabetically by business name. Revoked providers remain available for historical lookup when include_revoked is enabled. The hosted service runs in Sydney on Google Cloud Run and may take a few seconds to respond after being idle because it scales to zero.

Setup and configuration

For the hosted version, configure an MCP client with the endpoint URL; no environment variables or sign-up are required. The repository also supports local execution with Python 3.10 or newer. Install the requirements, then run the server against its bundled sample data over stdio. A locally built full feed can be supplied with --data.

The local server can also use streamable HTTP. Its port comes from PORT, defaulting to 8000. Building a current local feed involves downloading the Commission's CSV, running the ingestion script, validating the generated JSONL file, and starting the server with that file. The download may require a browser-style user agent.

Tools and capabilities

NDIS Provider Feed MCP server provides three tools:

• search_providers accepts optional suburb, state, postcode, service_category, include_revoked, and limit arguments. Suburb matching is case-insensitive and partial; state and postcode matching are exact. Service categories must match one of the values returned by list_service_categories. The default result limit is 20.
• get_provider returns the full record for a provider identified by ABN.
• list_service_categories returns the 38 valid NDIS registration groups for use in searches.

Provider records group outlets by ABN and preserve service categories found across those outlets. Each record includes a last_verified timestamp showing when it was last retrieved from the register.

Limitations and notes

The location filters describe head-office location, not the places where a provider delivers services. Outlet addresses are incomplete in the source, and service areas are not published. The feed covers registered providers only; an absent provider is not necessarily illegitimate, since unregistered providers may serve plan-managed and self-managed participants.

The register does not provide reviews, availability, pricing, or specialisations. Websites are missing for roughly a quarter of providers. Registration conditions are represented only as a boolean because the source text is lengthy and inconsistently encoded. Results are alphabetical rather than relevance-ranked.

The feed can lag the official register by up to a week. Confirm registration status with the official register before relying on it for a decision. Sole-trader records with an individual entity type omit the personal legal name and show only the trading name. Tool calls log timestamps, session IDs, client user agents, and search parameters; standard Cloud Run logs also record IP addresses, and logs are deleted after 30 days.