Research

Research is an async task for questions one search can't answer. Claude Haiku 5.5 plans Google queries for your input, the pages they find are read and ranked, a decision model (OpenAI's Decisions API) checks whether the notes already cover the task and drops off-topic sources, Claude reviews the gaps and plans follow-up rounds, then writes a Markdown report that cites numbered sources. With output_schema the result also has a JSON value in output.

Effort: lite runs 3 first-round queries in 1 round and reads 2 pages per query, for 30 credits; standard runs 5 queries over 2 rounds with 3 pages each, for 75; deep runs 8 queries over 3 rounds with 4 pages each, for 200. See the Deep research guide.

POST /v1/research returns 202 with a task. Poll GET /v1/research/{id} (progress while it runs, result when it ends; kept 24 hours) or receive the signed research.completed webhook. DELETE /v1/research/{id} cancels it: a queued task is refunded at once, a running one stops at its next round and is charged 1 credit per search already run.

The flat price is reserved when the task is queued and charged only when the report is written; research that fails is refunded in full.

Request body

Example request

Authenticate with Authorization: Bearer $SERPKITE_API_KEY (GET requests may pass ?api_key= instead). The official SDKs read SERPKITE_API_KEY for you. See Authentication.

curl https://api.serpkite.com/v1/research \
  -H "Authorization: Bearer $SERPKITE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input":"compare espresso machines under $500 for a small office","effort":"standard"}'

Response fields

Every billed response also carries credit and latency headers (X-Credits-Used, X-Credits-Remaining, X-Request-Id…).

Example response

{
  "id": "01926a3e-8c4d-7e1f-a2b3-c4d5e6f7a8b9",
  "kind": "research",
  "status": "completed",
  "created_at": "2026-10-24T12:00:00Z",
  "started_at": "2026-10-24T12:00:01Z",
  "completed_at": "2026-10-24T12:01:38Z",
  "credits_reserved": 75,
  "credits_used": 75,
  "webhook_status": null,
  "progress": {
    "round": 2,
    "rounds": 2,
    "queries": 8,
    "sources": 14,
    "step": "writing"
  },
  "result": {
    "input": "compare espresso machines under $500 for a small office",
    "effort": "standard",
    "report": "## Summary\n\nFor a small office, a heat-exchanger machine with a 2-litre tank is the safest pick under $500 [1][3]…",
    "citations": [
      {
        "index": 1,
        "title": "The best espresso machines under $500",
        "link": "https://www.example.com/best-espresso-under-500",
        "domain": "example.com"
      }
    ],
    "sources": [
      {
        "index": 1,
        "title": "The best espresso machines under $500",
        "link": "https://www.example.com/best-espresso-under-500",
        "domain": "example.com",
        "query": "best espresso machine under $500 office"
      }
    ],
    "queries": [
      "best espresso machine under $500 office",
      "office espresso machine tank size"
    ],
    "stats": {
      "searches": 8,
      "sources": 14,
      "rounds": 2
    },
    "llm": {
      "provider": "anthropic",
      "model": "claude-haiku-5-5",
      "input_tokens": 41200,
      "output_tokens": 1650
    }
  },
  "error": null
}

Errors

Errors use one shape: {"error":{"code","message","request_id"}}. Errors are never billed. Full list in Errors.

Notes

• Research tasks share the per-account task limit with crawls: at most 20 queued or running at once (429 beyond that).
• The input, the planned queries and the page text are sent to Anthropic (writing) and OpenAI (decision checks) under their API terms, which exclude training on API data. See Privacy and data retention.
• The remote MCP server has research and research_result tools.

Related

• Deep research guide
• Answer
• Webhooks
• Research API overview and pricing
• OpenAPI spec