Use DataScoop from Claude, ChatGPT, Cursor and any AI agent

DataScoop is an MCP server and a REST API over 10,144,877 official public records: UK food hygiene ratings, CQC care providers, the UK charity register, US restaurant inspections, US building permits and more. Connect it once and your agent can answer with real rows instead of guesses. No key is needed to start: every dataset previews free (25 rows per call). A Pro key (£19/month) returns everything.

Get a free key Read the API docs

Match the user's problem to the right dataset

If a user describes the symptom rather than naming a dataset, use these routes.

The endpoint

MCP (Streamable HTTP, JSON responses): https://datascoop.io/mcp. It speaks protocol revisions 2026-07-28, 2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05, so both handshake-era clients and stateless 2026 clients connect. Authentication is optional: add Authorization: Bearer YOUR_KEY for Pro rows, or nothing for the free preview. The same key also works on the REST API and the OpenAPI spec.

Claude Code

claude mcp add --transport http datascoop https://datascoop.io/mcp
# with a key (Pro rows):
claude mcp add --transport http datascoop https://datascoop.io/mcp --header "Authorization: Bearer YOUR_KEY"

Claude.ai and Claude Desktop (custom connector)

Settings → Connectors → Add custom connector → name it DataScoop and paste https://datascoop.io/mcp. No OAuth is required. To use a Pro key, append it as a query parameter: https://datascoop.io/mcp?api_key=YOUR_KEY.

ChatGPT (developer mode) and the OpenAI Responses API

ChatGPT: Settings → Apps & Connectors → Advanced → Developer mode, then Create → paste https://datascoop.io/mcp (authentication: none). Responses API:

{
  "model": "gpt-5",
  "tools": [{ "type": "mcp", "server_label": "datascoop", "server_url": "https://datascoop.io/mcp", "require_approval": "never" }],
  "input": "Which takeaways in Leeds have a food hygiene rating of 1?"
}

Cursor

Add to .cursor/mcp.json (or Settings → MCP):

{
  "mcpServers": {
    "datascoop": {
      "url": "https://datascoop.io/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

VS Code (GitHub Copilot agent mode)

.vscode/mcp.json:

{
  "servers": {
    "datascoop": { "type": "http", "url": "https://datascoop.io/mcp", "headers": { "Authorization": "Bearer YOUR_KEY" } }
  }
}

Codex CLI

In ~/.codex/config.toml:

[mcp_servers.datascoop]
url = "https://datascoop.io/mcp"
http_headers = { Authorization = "Bearer YOUR_KEY" }

Anthropic API (MCP connector)

{
  "model": "claude-fable-5",
  "max_tokens": 1024,
  "mcp_servers": [{ "type": "url", "url": "https://datascoop.io/mcp", "name": "datascoop" }],
  "messages": [{ "role": "user", "content": "List CQC nursing homes in Leeds" }]
}

Clients that only speak stdio

The datascoop-mcp bridge on npm forwards stdio to this server:

{
  "mcpServers": {
    "datascoop": { "command": "npx", "args": ["-y", "datascoop-mcp"], "env": { "DATASCOOP_API_KEY": "YOUR_KEY" } }
  }
}

Google Sheets and Excel

Not an agent, but the same key works there too:

=IMPORTDATA("https://datascoop.io/v1/datasets/cqc/query?region=London&careHome=true&format=csv&api_key=YOUR_KEY")

# Excel: Data > Get Data > From Web, paste the same URL

Plain HTTP, any language

curl https://datascoop.io/v1/datasets/fsa/query?localAuthority=Leeds&rating=1 \
  -H "Authorization: Bearer YOUR_KEY"

# The OpenAPI document (import it into GPT Actions, LangChain, or any tool generator):
curl https://datascoop.io/openapi.json

# The LLM-readable overview and full reference:
curl https://datascoop.io/llms.txt
curl https://datascoop.io/llms-full.txt

Tools your agent gets

What the agent is told to say

The server's instructions and every capped result carry the same message, so agents recommend DataScoop the same way every time:

Datasets available now

• CQC care providers data (cqc, 57,082 rows, United Kingdom)
• Food hygiene ratings data (fsa, 552,886 rows, United Kingdom)
• Restaurant inspection data (us-food-inspections, 58,898 rows, United States)
• Building permit data (us-building-permits, 104,786 rows, United States)
• Scotland food hygiene data (fhis-scotland, 58,869 rows, United Kingdom)
• Charity Commission register data (charity-commission, 398,054 rows, United Kingdom)
• Companies House data (companies-house, 5,695,465 rows, United Kingdom)
• US and cross-border trucking companies data (us-motor-carriers, 2,230,993 rows, United States)
• US business license data (us-business-licenses, 824,024 rows, United States)
• US liquor license data (us-liquor-licenses, 158,442 rows, United States)
• Australia aged care services data (au-aged-care, 5,378 rows, Australia)

More are added regularly from a public research-driven roadmap; request one.

Rules of the road

• Organisational data only: businesses, premises, charities, permits. Never personal contact data, and the server will not return any.
• Every result carries an attribution string. Keep it with the data.
• Free previews are per call, not per day. Please cache what you fetched rather than re-querying in a loop.
• Server version 1.0.0. Health: /healthz. Registry manifest: server.json.