Skip to main content

​
MCP server

The remote MCP server (Streamable HTTP) has one tool per endpoint, plus create_job, get_job, and get_job_result for async jobs. Every data tool takes a max_credits argument, and each tool’s description summarizes how it is priced. Two ways to connect:
  • OAuth. Add the URL as a custom connector in claude.ai, ChatGPT, or any client that supports MCP OAuth. You sign in to the console and approve one project, the scopes, and a daily or monthly credit budget. You can revoke access from the console at any time.
  • API key. Clients that support custom headers can send Authorization: Bearer hdk_sk_... from their secure configuration. Never put a key in a prompt.

​
Agent keys

Create a key with the Agent preset for any script or agent: it comes with 5 requests per second, a budget of 10,000 credits per UTC day, a 2,000-credit cap per request and a 30-day expiry. Also send X-Max-Credits on every request. See API keys.

​
Docs for models

https://data.hyperdash.com/llms.txt lists every dataset and endpoint with its limits and scope, in a format models read well. The full OpenAPI spec is at https://data.hyperdash.com/openapi.json. Errors carry a stable code, and many include a hint with the fix. Most 429 and 503 responses include Retry-After, so agents know when to retry. See Errors.

​
Example instructions