get_link_summary
Read per-website link counts by status (live, missing, couldn’t check, paused, not checked yet), last checked, and links found, back and lost over the last 1-90 days (found includes the links that came back; back counts those alone). The same numbers the console shows on Home, Reports and the website picker.
This command name is absent from deployment evidence dated 2026-09-20. Hosted availability for this current definition is unverified; inspect your connected catalog before invoking it. Verify the command and its input schema in the selected environment.
This read does not start a verification job.
Example request
Connect through MCP, CLI, or HTTP. Replace example identifiers with records from your workspace.Returned result
Illustrative data validated against the documented response schema. IDs and dates are examples, not a live account capture. MCP returns this data instructuredContent and in a text content block; generic HTTP and CLI return the JSON result directly.
Full response example
Full response example
Input fields
Omit optional fields when you do not want to supply them. Null is accepted only where listed. Unknown input properties are rejected.Validation and omitted values
Omit projectId to read every website allowed by the credential, or supply one accessible saved website id. days accepts a whole number from 1 to 90 and defaults to 30. The period covers that many UTC calendar days, including today, and ends at as_of. Each monitored link belongs to one status. Paused checks take priority; live means present or changed, and missing requires confirmed missing. An inconclusive, incomplete, unavailable or suspected-missing check is counted as couldnt_check, even when an older check confirmed the link. An unknown link with no check attempt is not_checked_yet. last_checked_at is the latest active check attempt, including an inconclusive attempt; it excludes paused links and does not establish presence. found counts retained placement-acquired and recovered events; back is the recovered subset, so do not add found and back together. lost counts retained placement-lost events. Without events:read, found, back, lost, daily and changes_complete are null because changes are unknown, not zero. Website rows are capped at 1,000 and websites_truncated reports omitted rows; totals still cover the entire readable selection. Change counts examine at most the latest 100,000 workspace event sequence positions. A retention floor or scan cap makes changes_complete false. Even a complete quiet period says nothing about unobserved links outside the tracked dataset. Reads saved monitoring records only. It does not fetch publisher pages, queue checks, run classification or spend supplier or model credits. as_of identifies the response time; the summary is not a paginated export or an immutable snapshot of all website records.Defaults when omitted
Output fields
Fields inside optional or nullable parents apply only when that parent exists. Common schema conventions explain evidence states, empty lists and extensions.schema_version(number, required): must equal 1workspace_id(string, required): Workspace that owns this record.project_id(string / null, required): Project that owns this record.as_of(string, required).period(object, required).totals(object, required).websites(array, required).websites_truncated(boolean, required).daily(array / null, required).changes_complete(boolean / null, required).statuses(array, required).
All fields and nested objects
All fields and nested objects
Errors and retries
Follow the error recovery before retrying this read. Preserve the selected workspace, project, filters and feed. Recovering a read does not require an idempotency key or another verification job. Operation-specific errors includeINVALID_INPUT, PROJECT_NOT_FOUND, WORKSPACE_ACCESS_DENIED, INSUFFICIENT_SCOPE.
See error recovery for status, scope, cooldown, cursor and retry handling. Unknown observations are result data and do not establish loss.