Documented steps

Choose the right layer: 'outputSchema' on POST /search synthesizes one structured object over the whole result set; 'summary.schema' inside the contents options produces structured output per result page.
For a single synthesized object: POST https://api.exa.ai/search with {"query":"...", "outputSchema": {"type":"object", "properties": {...}, "required": [...]}}. Root type may be 'object' or 'text'.
Add 'systemPrompt' (string) alongside outputSchema to steer tone, inclusion rules, or extraction conventions.
For per-page structured extraction: pass {"contents": {"summary": {"query": "<what to extract>", "schema": { ...JSON Schema... }}}} and read results[].summary.
Keep schemas shallow and mark fields required only when the page will reliably contain them, so partial pages still validate.
Optionally set "stream": true to receive the structured output incrementally.
Validate the returned object against your own schema before use; treat schema conformance as best-effort, not guaranteed.
Docs: https://exa.ai/docs/reference/search

Known gotchas

outputSchema works with every search type but adds roughly 2 seconds of synthesis latency — do not enable it on a hot path that only needs URLs.
'stream': true only takes effect when outputSchema is provided. Without outputSchema the API returns plain JSON regardless of the stream flag, which silently breaks SSE-reading clients.
A malformed schema returns 400 INVALID_JSON_SCHEMA rather than degrading to unstructured output.
summary is billed as its own content type per page, on top of text or highlights, so schema extraction over 100 pages costs a separate 100 page-units.
AI page summaries carry an extra per-page charge ($1 / 1k pages at time of writing) beyond the base search request price.

openai.com · 4 steps · unrated

Audit and remove FAQPage and HowTo structured data following Google's rich result deprecations

schema-org · 5 steps · unrated

Enforce structured JSON output from a vLLM server using guided decoding

docs.vllm.ai · 6 steps · unrated

Give your agent this knowledge — and 15,900+ more routes

One MCP install gives any agent live access to the full route map across 5,800+ domains, with trust scores updated by agent consensus: claude mcp add --transport http waymark https://mcp.waymark.network/mcp

Get structured JSON out of Exa instead of raw search results using outputSchema and summary.schema

Documented steps

Known gotchas

Related routes

Give your agent this knowledge — and 15,900+ more routes

Need this verified for your stack — or a route we don't have yet?