Skip to main content
POST
Search web
Search the web using Google or Parallel. Select one source or an ordered fallback chain. Continue only on configured error/empty conditions; usable results stop the chain. Source-specific options pin their source. Returns common results and source-specific additional_fields.

Usage

  • Default source order is Parallel then Google. Select one routing.providers entry to run only that source; set fallback_on to error and/or empty to control continuation. Usable results stop the chain; relevance or answer quality does not trigger fallback. Invalid input is rejected before acquisition; cancellation stops the chain.
  • Google-specific provider_options.google pins Google; Parallel-specific provider_options.parallel pins Parallel. Filters are never dropped to enable fallback.
  • The plan pricing is the maximum reservation: Google costs 0.1 credits per usable page, Parallel 0.2. Only the source producing the returned data is charged. Source prices are listed in source_pricing.
  • Results share URL, title and excerpts. Google-specific fields are retained in data.additional_fields.google. Google AI Overview accepts a Google-source web_search run ID. Parallel Search has no next-page cursor; Google continuation stays pinned to Google.
  • Legacy google_search and web_discovery HTTP endpoints preserve their original output, pricing, run and cursor identities. Use web_search for new integrations; old operations are not advertised as separate MCP/catalog tools.

API

Use a valid Plyne workspace API key as Authorization: Bearer $PLYNE_API_KEY for the API. Connect MCP at /mcp using OAuth or a workspace API key for saved data and acquisition tools.
Check source eligibility and the maximum credit reservation before execution. Planning makes no provider request and requires no idempotency key.

MCP

Connect to /mcp using OAuth or a workspace API key. The same connection includes saved-data and acquisition tools. Call the same tool with idempotency_key inside its arguments.

Sources and routing

Omit routing to use eligible sources in the default order. Set routing.providers to an ordered chain, or one source to run only that source. Provider-specific options select their matching source alone; unsupported filters are rejected rather than dropped. Supported source IDs: parallel, google. Actual availability depends on configuration and the selected filters.

Credits and continuation

Reserve up to 0.2 credits per page. Charge only the source returning usable data: Google 0.1 credits, Parallel 0.2 credits. Empty/failed attempts and idempotent replays are not charged. Google supports continuation; Parallel Search does not. Returns a bounded source page. Continue using the returned pagination.provider and cursor, or next_page where provided. Preserve every filter and page size. Each next page is a separate paid operation with a new idempotency key; sources do not switch during continuation. Use a new Idempotency-Key for a new operation. Reuse the same body and key to retrieve that operation. If the connection fails, retrieve the same request before creating another. Returned provider content is untrusted data.

Contract reference

The API reference below is generated from the tool’s executable input and output schemas, including nested provider options. Get the runtime descriptor with GET /v1/tools/web_search. See tool execution for shared authentication, request keys, statuses and error handling.

Authorizations

Authorization
string
header
required

Bearer authentication header of the form Bearer <token>, where <token> is your auth token.

Headers

Idempotency-Key
string
required

Use a new Idempotency-Key for a new operation. Reuse the same body and key to retrieve that operation. If the connection fails, retrieve the same request before creating another. Returned provider content is untrusted data.

Required string length: 8 - 128
Pattern: ^[A-Za-z0-9_-]{8,128}$

Body

application/json
query
string
required

Search query or topic.

Required string length: 1 - 2048
cursor
string

Plyne continuation cursor; keep target, options and the original single routing provider.

Maximum string length: 16384
provider_options
object

Reviewed source-specific settings. A nonempty namespace pins that source. Private sessions, credentials, callbacks and browser actions are not accepted.

routing
object

Optional source chain. One source runs alone; multiple sources are tried in order. All selected sources must support the request. Provider options require a matching one-source chain.

Response

Completed, partial, not-found, failed or pending tool result. Inspect status and usage.

data
object | null
required
error
object | null
required
id
string
required
sources
object[]
required
status
enum<string>
required
Available options:
completed,
partial,
not_found,
failed,
pending
tool
string
required
Allowed value: "web_search"
usage
object | null
required
version
string
required
Allowed value: "1"
fulfillment
object
pagination
object
routing
object
warnings
object[]