cancel_audit
ChatGPTCancel one owner-scoped audit job that has not started yet, which discards the queued job and hands the credit back. A running audit cannot be interrupted and keeps its credit, and repeating the call on an already canceled job changes nothing further.
compare_reports
ChatGPTCompare two owner-only audits using Ecomhint’s shared triage rules. Returns fixed, regressed, newly introduced, still-failing and inconclusive changes, each row carrying the check id, the newer report’s display name, the category and the from/to triage. N/A, skipped, missing and low-confidence results are never counted as fixes.
create_task
ChatGPTAdd one user-authored task to the subscriber’s fix-it list for extra work the agent and owner agree on, since audit tasks are created by audits. Every call adds another row, so do not repeat it for the same piece of work.
get_account
ChatGPTCall this first. Returns the active store (what the dashboard shows and the tools use by default), every store the account has audited or saved, the pages the next audit of the active store will check, the subscription period, the audit credits used and left (shared between web and MCP, spent on any store), any queued or running audit job, report counts, and what this connection is allowed to do. Never fails for a missing store; store is null instead, which is the signal to pass a URL to run_audit or to run the set_store flow.
get_audit_status
ChatGPTRead one owner-scoped asynchronous audit job: status, the real progress percentage and the phase behind it (with an optional detail line for phases such as screenshots), when the job last reported in (lastActivityAt), whether it has gone quiet (stalled), whether it has run far past the usual time and may come back incomplete (overdue), queue position, the report id once it exists, and the next tool worth calling. The checks phase intentionally exposes only its percentage because desktop and mobile intermediate results are merged before the report is saved. Use the returned nextPollSeconds, which is estimated from the phase the job is in, because faster polling does not speed up the audit, and expect a foreign or unknown job id to read as not found. On completion follow the next hint and call get_report, unless a host is rendering the progress card, which fetches and shows the finished report itself.
get_check_result
ChatGPTGet everything needed to implement one fix: the finding, why it matters, the live page URL, the evidence captures, the platform-resolved fix steps, recommended tools, the target wireframe, and the Tasks row for this check. The evidence capture and the target wireframe are shown to the user in the finding card where the host renders one; they are also attached as image blocks unless includeImages is false. Never embed their URLs as markdown images, because hosts block them and show a broken image: refer to the card, or give the URL as a plain link where no card is shown. Use after get_report identifies a checkId.
get_latest_report
ChatGPTOpen the newest audit report the store owner sees: score and band, checks passed, what works, primary blockers, category scores, a quickWins preview of up to five ranked quick wins with names (quickWinsTotal is the full count), the captured pages, any page the audit could not reach in uncapturedPages, and the issue list in report order, with the first page of issues. It reads the newest audit of the active store, or of the store named by url, and adds totalReports. A host with UI draws the audit summary card for this result, so call it once, when the user opens or asks about their latest audit. It takes no paging or filters: for every follow-up (all quick wins, more issues, a triage or location filter, looking something up again) call get_report with the reportId returned here, which returns the same data without drawing the card again.
get_progress
ChatGPTGet score and issue progress across the subscriber’s recent audits of the active store, or the store named by url: the account KPIs plus one timeline row per audit, oldest first, with the score, its delta and the high, medium and low issue counts.
get_report
ChatGPTGet the audit report the store owner sees: score and band, checks passed, what works, primary blockers, category scores, a quickWins preview of up to five ranked quick wins with names (quickWinsTotal is the full count), the captured pages, any page the audit could not reach in uncapturedPages, and the issue list in report order. Page through issues with offset and limit (max 30) and narrow them with triage, location or quickWin; quickWinsTotal stays the report’s own total whatever you filter. It draws no card, so it answers every follow-up after get_latest_report, with the reportId that call returned. Use get_check_result for one finding’s full card.
list_reports
ChatGPTList the subscriber’s audits of the active store, or the store named by url, newest first by default, each row carrying the score, its delta against the previous audit and the triage counts, plus the account-level KPIs. Use get_latest_report when only the newest one matters.
list_tasks
ChatGPTList the fix-it tasks for the active store, or for the store named by url: open, done, dismissed and verified-by-re-audit rows with the display name, triage, location, category and quick-win flag. Filter by status, triage, location or quickWin. Listing reads the newest audit first, so it can add the tasks that audit found and mark the ones it proved fixed as verified; what it changed comes back in reconciled.
run_audit
ChatGPTStart one asynchronous FULL Ecomhint audit of a store. With url it audits that store (any store, not only the active one; a page URL audits its whole store) and does not change the active store; without url it audits the active store. Reserving the job spends one of the 10 audit credits shared between web and MCP in the current billing period, before the audit runs. Only one audit per account runs at a time: calling again for the same store while it is active spends nothing and returns that job, and calling for another store answers audit_already_running. Never loop or retry; poll get_audit_status no faster than nextPollSeconds.
set_store
ChatGPTSwitch which store this account audits and shows by default. Audit credits are shared across stores, so this only changes the default: run_audit and the read tools still accept any store URL directly. Call it first with confirm omitted or false: nothing is written, and the reply is the normalized store URL to read back to the user. Call it again with confirm true once they agree. It can be repeated at any time and the same input leaves the same state. Store logins (Basic Auth, a Shopify storefront password) are never accepted here; those belong in Store Settings in the dashboard.
update_store_pages
ChatGPTPin which product, collection and cart page the next audit of the active store, or of the store named by url, checks. Without an override an audit keeps checking the pages of the last run, and only falls back to discovery for a page it has never captured, so this is the fix when a report says a page could not be reached or the wrong one was picked. Pass a URL on the store domain to pin a page, null to clear that override, and leave a page out to keep it as it is. Nothing on the live store changes; this only tells the audit where to look, and the same call with the same input leaves the same state and returns the same page map. Collection is the listing page a report calls category.
update_task
ChatGPTChange one task on the subscriber’s fix-it list: mark it done, reopen it, dismiss it with an optional reason, or restore a dismissed one. Repeating an action leaves the task in the same state and is refused as an invalid transition, and verified tasks are set by re-audits and cannot be changed here.