Query public records from your AI tools

DataScoop provides 10,174,536 records across 11 datasets through MCP and a REST API. Find a dataset, check its coverage and fields, then filter the records for your task. No key is needed for a free query of up to 25 rows.

See worked examples Read the API docs

Start with a task

Each example includes a prompt, exact filters, code and a dated sample. The sample illustrates the output; query the service to check its current saved import.

• Build an exact-area care-home directory
• Research restaurants by council and business type
• Find carriers by location, operation and fleet size

Access and limits

Anonymous and free-key queries return up to 25 rows per call and support offset. Pro costs $9.99 once for 12 months, with no automatic renewal. It supports up to 1,000 rows per MCP call or 10,000 rows per REST JSON page, plus up to 10 CSV downloads per account per day, resetting at midnight UTC. A separate full CSV costs $1.99 without an account. See prices and fair-use details.

A page with truncated=true is incomplete. Use offset to retrieve the next page, including on the free plan. A full Pro CSV uses format=csv without limit; each started CSV request uses the daily allowance.

Connect a client

Endpoint: https://datascoop.io/mcp (Streamable HTTP, JSON responses). Configuration and authentication vary by client. Start without a key; add an existing DataScoop key only where the client supports a custom Bearer header. DataScoop does not currently provide an OAuth account-linking flow. These instructions were checked against the linked vendor documentation on 21 October 2026.

Claude Code

Add the public endpoint, then check its status with /mcp. Official Claude Code instructions.

claude mcp add --transport http datascoop https://datascoop.io/mcp

For an existing key, Claude Code supports an environment reference in a private or project .mcp.json. Set DATASCOOP_API_KEY in the environment that starts the client; keep the actual key out of the file.

{
  "mcpServers": {
    "datascoop": {
      "type": "http",
      "url": "https://datascoop.io/mcp",
      "headers": {
        "Authorization": "Bearer ${DATASCOOP_API_KEY}"
      }
    }
  }
}

Codex

Add this to ~/.codex/config.toml. Omit the optional token setting for free access. For Pro, set the named environment variable before starting Codex. Official MCP configuration.

[mcp_servers.datascoop]
url = "https://datascoop.io/mcp"
# Optional: existing DataScoop key from the client environment
bearer_token_env_var = "DATASCOOP_API_KEY"

Cursor

Put this free-access configuration in ~/.cursor/mcp.json or the project’s .cursor/mcp.json. Cursor also supports an Authorization entry in headers; keep a real key in private configuration, not a shared project file. Official Cursor instructions.

{
  "mcpServers": {
    "datascoop": {
      "url": "https://datascoop.io/mcp"
    }
  }
}

VS Code with GitHub Copilot

For free access, add this to .vscode/mcp.json and enable the server in the tool picker. Documented VS Code configuration format.

{
  "servers": {
    "datascoop": {
      "type": "http",
      "url": "https://datascoop.io/mcp"
    }
  }
}

ChatGPT

Where custom MCP connections are enabled, follow OpenAI’s custom MCP instructions: add the endpoint with No authentication for public queries, install the resulting personal connection and select it in a chat. Workspace permissions apply. This setup is not a public directory listing.

ChatGPT’s documented account-linking route uses OAuth; it does not accept a customer API key as a custom authentication method. DataScoop has no OAuth flow, so Pro account linking is not supported here. Do not put a key in the connection URL. See OpenAI’s authentication requirements. No in-chat checkout is offered by this setup.

Developer APIs

For the OpenAI Responses API, add this MCP entry to the tools array of a request using a model that supports remote MCP. Approval settings remain your application’s choice. See the official MCP guide for authentication and approval handling.

{
  "type": "mcp",
  "server_label": "datascoop",
  "server_url": "https://datascoop.io/mcp"
}

The Anthropic Messages API has a separate MCP connector with its own beta header and tool configuration. Follow that guide for the SDK version you use; the client configurations above are not Messages API request bodies.

Clients that launch a local stdio process

The published datascoop-mcp npm bridge forwards local JSON-RPC messages to the hosted service. Node.js 18 or later is required. For an existing key, pass DATASCOOP_API_KEY to the process using the client’s private environment configuration.

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

Make the query explainable

Discover the source with search_datasets or list_datasets, then read get_dataset_schema. Set strict_filters: true on MCP queries or strict_filters=1 on REST to reject unsupported or invalid predicates. Without strict mode, check ignored_filters in MCP or ignored in REST before using a result.

Keep applied_filters, provenance and attribution with your answer. provenance.imported_at records a successful DataScoop import; it is not the publisher’s release date or a record’s update time. Compare it across pages of a long result. Preserve source dates where available and check missing fields before drawing conclusions.

Plain HTTP

curl "https://datascoop.io/v1/datasets/fsa/query?localAuthorityExact=Leeds&businessTypeExact=Restaurant%2FCafe%2FCanteen&strict_filters=1&limit=5"

# Machine-readable reference
curl "https://datascoop.io/openapi.json"
curl "https://datascoop.io/llms-full.txt"

Tools available

Other supported tasks

Datasets available now

• Download CQC care-provider data (cqc, 57,118 rows, England)
• Food business lists and hygiene ratings data (fsa, 555,523 rows, England, Wales & Northern Ireland)
• Restaurant inspection data (us-food-inspections, 60,365 rows, New York City, Chicago, New York State, Austin)
• Building permit data (us-building-permits, 107,128 rows, New York City, Chicago, Los Angeles, Austin, San Francisco, Seattle)
• Scotland food hygiene data (fhis-scotland, 58,934 rows, Scotland)
• Charity Commission register data (charity-commission, 398,905 rows, England & Wales)
• Companies House data (companies-house, 5,704,711 rows, United Kingdom)
• Build a US carrier list by state and fleet size (us-motor-carriers, 2,242,833 rows, United States, Canada and Mexico)
• US business license data (us-business-licenses, 823,837 rows, Chicago, New York City, San Francisco, Los Angeles)
• US liquor license data (us-liquor-licenses, 159,804 rows, New York, Texas, Missouri, Colorado)
• Australia aged care services data (au-aged-care, 5,378 rows, Australia)

Need a different source? Tell us the task, country and official publisher.

Using and citing the records

Fields marked personal are withheld. Published business names, addresses and phone numbers may still identify people, including sole traders. Retain source attribution and dates, follow each dataset’s reuse terms, and assess privacy and marketing obligations for your intended use. Read the methodology for coverage and missing-field limits.

Cache results where appropriate and follow the API fair-use terms. Server version 1.0.0; service health; registry manifest. Connecting a tool makes it available to that client; it does not guarantee recommendations in search answers.