> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.usescout.sh/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.usescout.sh/_mcp/server.

# Glossary

These are the words that turn up in the API reference and guides.
Skim once when you start, refer back when a field name does not match
what you expected.

## Core concepts

**Credit.** The billing unit. One credit covers one Google SERP fetch.
Two credits cover one page extract. Different SKUs are listed in the
`usage[]` array on every response. See [Pricing](/pricing).

**Objective.** A short natural-language sentence that tells Scout what
you actually want from a request. On `/v1/search` it re-ranks results
by relevance and drops off-topic rows. On `/v1/extract` it triggers
auto-discovery of subpages that match the objective. The field is free
text — `"play store URLs"`, `"pricing per plan"`, and `"funding amount
and investors"` are all valid.

**SKU.** A named line item on the `usage[]` array, e.g. `sku_search`,
`sku_objective_rerank`. Each is billed independently and only appears
when the feature was used.

## Endpoints, by shape

**Synchronous endpoint.** Returns the full response in one HTTP call.
Used by `/v1/search` (at `depth: instant`, `fast`, or `auto`),
`/v1/extract`, and `/v1/chat/completions`.

**Asynchronous endpoint.** Returns a `task_id` or `search_id` in under
a second, then runs the work in the background. Poll the run, stream
its events over SSE or WebSocket, or set a `webhook` URL to receive the
completed payload. Used by `/v1/task`, `/v1/search?depth=deep`,
`/v1/findall`, and `/v1/monitors`.

**Run.** A single execution of an asynchronous endpoint. Identified by
its `task_id`, `search_id`, `findall_id`, or `monitor_event_id`. Status
moves through `queued → running → completed | failed | cancelled`.

## Search-specific

**Depth.** Controls how hard Scout works for a search.

* `instant` — fewer rows, single SERP page, lowest latency.
* `fast` — single SERP page, ten rows per query.
* `auto` (default) — up to five SERP pages per query, paginated in
  parallel to satisfy `limit`.
* `deep` — async; planner-driven multi-query research, returns a
  citation-backed report.

**Excerpt.** A short verbatim slice from a search result. Surfaced
under `excerpts[]` on each row, capped by
`advanced_settings.excerpt_settings.max_chars_per_result`.

**Highlight.** A focused excerpt picked by an LLM based on the
objective. Opt in via `advanced_settings.contents.highlights`.

**Summary.** A free-text or structured-JSON summary of a result page,
written by an LLM. Opt in via `advanced_settings.contents.summary`.
Supports OpenAI's `response_format` shape for JSON-schema output.

## Task-specific

**Processor.** The model and effort tier the agent uses. Pick when you
create a task.

* `base` — Claude Haiku, no thinking, capped at three tool calls.
  Fastest and cheapest.
* `core` — Claude Sonnet with adaptive thinking, up to eight tool
  calls. The default.
* `ultra` — Claude Opus with high effort, up to fifteen tool calls.
  Best when correctness matters more than cost.
* `research` — multi-step planner → fan-out search → fetch → cited
  report. Used for genuine open-ended research questions.

**Findings.** Per-sub-query answers on a research task. Each carries an
answer, a confidence label, and the indices of the sources that back
the answer.

**Basis.** Per-field citation evidence on a task's structured output.
For each field on the `output` object, `basis[]` lists the URLs Scout
actually read to populate it, plus a one-line reason.

## Monitor-specific

**Cadence.** How often a monitor runs. Either a friendly name (`hourly`,
`daily`, `weekly`) or a five-field cron expression.

**Mode.** `diff` (default) only emits an event when content changes
between runs. `snapshot` emits on every run.

**Filter prompt.** Optional natural-language predicate. The monitor
runs as scheduled, then asks Claude whether the result satisfies the
predicate. The event is marked `matched`, `skipped`, or `ok`
accordingly.

**Webhook signature.** When `webhook_secret` is set, outgoing webhook
calls include an `X-Scout-Signature: sha256=<hex>` header — the HMAC
of the raw request body. Use this to verify the payload came from
Scout. See [Webhooks](/webhooks).

## Common response fields

**`credits`.** Total billing units for the call.

**`usage[]`.** Per-SKU breakdown of the same number.

**`warnings`.** A list of soft issues that did not fail the request
(e.g. a domain in `exclude_domains` that did not match anything).
`null` when no warnings apply.

**`X-Cache`.** Response header — `HIT` if the result was served from
the SERP cache, `MISS` otherwise.

**`X-Elapsed-Ms`.** Response header — server-side processing time.