Radar Crawler REST API & MCP Tools

Call Radar's Crawler from your own code or AI agent via REST API or MCP tools — same crawl engine, pricing, and contact extraction as the dashboard.

Everything the Crawler does in the dashboard — Brand and Directory crawls, contact extraction, and the optional HTML/Markdown page output — is also available as a REST API and as MCP tools, so you can call it from your own code or point an AI agent at it directly.

Authentication

Create a key at Settings → API Keys in the dashboard. The plaintext key is shown exactly once — store it yourself, Radar only keeps a hash. Send it as a bearer token on every request:

Authorization: Bearer rdr_live_...

Rate limit: 60 requests/minute per key, shared across every Radar API/MCP operation. An invalid or revoked key returns 401; going over the limit returns 429 with a Retry-After header.

REST API

Base URL: https://radar.yaeris.com/api/v1

POST /crawl — start a crawl.

{
  "type": "BRAND",
  "urls": ["https://example.com"],
  "pageLimit": 30,
  "depthLimit": 2,
  "format": ["contacts", "markdown"]
}

• type: "BRAND" (crawl one or more business sites directly) or "DIRECTORY" (crawl a listing/index page and follow individual listings).

• urls: required, non-empty.

• pageLimit / depthLimit: optional (default 100 / 2).

• format: optional, defaults to ["contacts"]. Any combination of:

  • "contacts" — business name/phone/email/website (today's default behavior).

  • "html" — the raw fetched page HTML, capped at 300,000 characters (truncated with a trailing marker past that).

  • "markdown" — a readability-cleaned Markdown version of the page content, capped at 100,000 characters — the format an LLM/RAG pipeline typically wants, far cheaper in tokens than raw HTML. Falls back to converting the whole page if it isn't article-shaped enough for readability extraction to find a main content block (e.g. a directory listing page), so it's never returned empty just because a page isn't an article.

→ 202 { "jobId": "...", "status": "QUEUED" }

GET /crawl/:jobId — status and results.

{
  "jobId": "...",
  "type": "BRAND",
  "status": "DONE",
  "error": null,
  "createdAt": "...",
  "finishedAt": "...",
  "pages": [
    { "url": "...", "httpStatus": 200, "fetchStatus": "SUCCESS", "businessName": "...", "phone": "...", "email": "...", "websiteUrl": "...", "html": null, "markdown": "..." }
  ]
}

status is one of QUEUED / RUNNING / DONE / FAILED. A page's fetchStatus (SUCCESS / BLOCKED / FAILED / TIMEOUT) explains that specific page even when the overall job succeeded. html/markdown are null unless requested via format on the original POST /crawl call.

Charged 1 credit per page crawled, once the job finishes — plus 1 extra credit per page if format includes "html" and/or "markdown" (both admin-configurable in the dashboard's Admin → Pricing). Your balance is checked against the maximum possible cost (page limit × rate) before the crawl starts, so an empty balance returns a 402 immediately rather than the crawl starting and failing partway through — the actual charge afterward is capped by pages actually visited, never more.

Errors

Status	Meaning
400	Bad request — missing/invalid fields (message explains which).
401	Missing, malformed, or invalid Authorization header.
402	Insufficient credit balance for this operation.
404	The referenced job doesn't exist on this project.
429	Rate limit exceeded — check Retry-After.

Every error body is { "error": "..." }.

MCP server

Endpoint (Streamable HTTP, stateless): https://radar.yaeris.com/api/mcp

Same bearer-token auth as the REST API, sent as a header on the connection.

Claude Code (project-scoped .mcp.json, or claude mcp add):

{
  "mcpServers": {
    "radar": {
      "url": "https://radar.yaeris.com/api/mcp",
      "headers": { "Authorization": "Bearer rdr_live_..." }
    }
  }
}

Claude.ai / other MCP clients: add a remote MCP connector pointing at the same URL, with the same Authorization header.

Tool	Equivalent to
start_crawl	POST /crawl
get_crawl_job	GET /crawl/:jobId

Each tool wraps the identically-named REST operation above — same inputs, same validation, same credit charges. A typical agent calls start_crawl, then polls get_crawl_job until status is no longer QUEUED/RUNNING.

Dashboard-only for now: Bulk Crawl (crawling many starting URLs from a saved list in one go). The API/MCP tools above cover single and multi-URL Brand/Directory crawls, just not that saved-list workflow yet.