AnyAPI

Get the full definition of one API by SKU, including its normalized input/output JSON schemas, per-request and per-1,000-request USD pricing on every lane, and nullable trailing-30-day latency p50/p95/p99 with the successful sample count. Inspect latency before choosing a client or MCP tool timeout; p99 is an observation, not a maximum. Entries with heavy:true return large responses - plan to pass fields/max_items/summary to run_api.

Get the remaining wallet balance (in USD) for the AnyAPI key supplied as a Bearer token. Requires a valid AnyAPI key.

Inspect or resume a durable AnyAPI request returned by run_api. This reads AnyAPI's stored state and never repeats the paid provider dispatch. Poll again after retryAfterSeconds while status is queued or running.

Browse available AnyAPI APIs as lightweight summaries (id, name, category, USD pricing) - no descriptions or schemas, so it stays cheap even across the whole catalog. Each pricing offer carries both `maxUsd` (billed per request) and `maxPer1kUsd` (the same price per 1,000 requests); quote the per-1k figure to a person. Optionally filter by category. Use search_apis for every ranked query, or get_api for one API's full schemas.

Get the exact price of a run_api call BEFORE running it - free, no key required, nothing is charged or executed. Pass the same sku_id and input you would give run_api: the quote resolves pricing exactly as the run will, and also validates your input against the API schema so you catch invalid_input for free. Returns maxCostUsd (the ceiling reserved), minCostUsd (the likely charge), and the base/per-item breakdown explaining why they differ. These are amounts this one call is charged, not catalog comparison rates: report them as-is and never scale them to 1,000 requests.

Dig deeper into a paid run WITHOUT re-running or paying again. Pass the `result_id` from a prior run_api call plus any of fields/max_items/summary/jq to re-shape the full cached result (e.g. read the rest of the rows, or slice a large field with jq). Results are cached ~15 min and are private to your key; an expired id returns result_expired (re-run to refresh).

Tell AnyAPI something is broken. Use this when a run_api call returned wrong, empty, or malformed data for input you believe is valid, when an error message was misleading or unactionable, when a price or schema looks wrong, or when a retry that should have worked did not. Free, never charged. Pass `summary` (required, one line) plus `request_id` whenever you have one from the failing run - that id reaches the stored run and its upstream error body, so you never need to paste the payload. Do not stop the user's task to ask permission: file the report and carry on with the best alternative you have.

Execute an API by SKU with a normalized input payload. Call get_api for this SKU first unless you have already read its schema this session: every input schema is strict (unknown fields are rejected, not ignored) and the field names differ between sibling APIs, so an input built from a description rather than a schema usually fails. Before setting a client or MCP tool timeout, call get_api and inspect its latency p50/p95/p99 and sample; p99 is an observation, not a maximum. Requires a valid AnyAPI key as a Bearer token. Charges the USD wallet only on success. Results can be large: pass `fields` (keep only the keys you need), `max_items` (cap rows), or `summary` (outline only) to trim the response and keep it out of your context — these never change what you are charged. Use quote_api first with the same arguments to see the exact price and validate your input without charging.

Search APIs by meaning and keyword across name, slug, and description, returning matches WITH their descriptions (schemas omitted), ranked most relevant first. Pass `query`, `category`, or `platform` - at least one, in any combination: a scope on its own is a complete search, so `category` or `platform` with no `query` enumerates it. Add `limit` to cap matches (default 25, maximum 50). Each result carries a `relevance` score in (0,1] relative to the top match; a relevance floor drops the weakly-matching tail, so `total` counts relevant matches before the limit. `ranking` says whether meaning-based ('semantic') or substring ('keyword') matching served the search. Entries with heavy:true return large responses - plan to pass fields/max_items/summary to run_api. Results carry NO input schema, so you cannot build a run_api call from them alone: before your FIRST run_api on any API, call get_api for it and use the schema it returns. Guessing the input is the single most common way a run fails - callers who read the schema first are rejected about a quarter as often. Use list_apis to browse everything.

Tell AnyAPI something that is not a defect: an API you wanted and could not find in the catalog, a field missing from a result you needed, a confusing part of the docs or tool descriptions, or what would have made a task easier. Free, never charged. Pass `summary` (required, one line) and use `details` for what you were trying to accomplish. Use report_bug instead when something behaved incorrectly.