Approve was clicked but the return did not finish
Your choice may already have been recorded even if the browser did not reach the assistant or publishing page. Do not keep approving the old form. Use the Continue button on the return screen once; if the host says the request expired, start a fresh sign-in from the original assistant or app-publishing page.
The publisher's tool-discovery authorisation and each customer's connection use the signed-in ScanMySEO account, but are separate from verifying an organisation or publishing a directory listing. The permissions shown on each consent screen are the ones requested for that step.
Sign-in expired or the connection is no longer valid
Restart authorisation from the assistant and inspect the account and permissions. Do not paste a personal token into an OAuth secret field. Do not reuse a consumed callback URL. For renewable clients, a concurrent refresh or detected replay may require deliberate reconnection.
An action is unavailable
The reason matters. Permission, account allowance, selected websites, optional connection limits, operator availability and concurrency are separate checks. Reconnecting does not cure a depleted account or an intentionally disabled operator gate.
When an existing audit is still active, follow that operation. When a quote is stale, obtain a current quote only after confirming no operation was accepted. Do not assume that a request which timed out never ran.
An AI request is awaiting approval
Read the exact audit, revision and question or enhancement action on the approval page. Merely opening that page runs no provider work. Expiry, revocation, changed question or a changed credential can invalidate approval. An uncertain external attempt is not a free retry.
The report is incomplete or the component does not appear
Read the coverage and evidence limitations. Missing retained evidence cannot establish a fix. Use the structured/text report when an assistant cannot render Audit Review. For a UI test, enable the host's normal content security enforcement and retain a redacted result.
Get help
Use Contact and support. Include the relevant audit or operation reference and a redacted error when necessary. Do not send access/refresh tokens, authorisation codes, cookies, full OAuth URLs or private conversation contents. Technical setup is an intentionally advanced path, not the normal customer connection journey.
Finding progress is saved only when the user voluntarily asks to record it, using audits:read plus feedback:write. Applied is a client-reported claim, not a verified fix. Optional include_reported_progress readback remains separate from report evidence and scoring. Saved-audit comparison does not start a scan or consume credits. Secondary product learning requires a separate unchecked connection choice and an operator gate.
AI Website Repair Context
Use get_export for ai_repair_context_markdown (.md, assistant upload) or ai_repair_context (.json, agents). Both represent the same authorized saved evidence under scanmyseo.ai_website_repair_context.v1. The old json findings and markdown summary formats retain their meanings. Metadata does not compile a file. Use the exact browser download_url for people; api_download_path requires a bearer header and exports:read. Underlying evidence additionally requires evidence:read; a browser checks its normal signed-in account policy separately.
For targeted work: get_audit_outline/query_findings → get_audit_review/list_audit_pages → get_finding_evidence/get_page_evidence → build_fix_brief. Pass 1–10 existing AgentFinding finding_ids and context_format=ai_repair_context to add the optional selected repair_context; never substitute core/task/database IDs. The default brief stays unchanged. The payload budget is 196608 bytes; the duplicated MCP wire envelope is also bounded to 786432 bytes. If too large, select fewer findings or request the full file.
Confirm current site state and intended behavior, map evidence to the authorized project, review changes and test functional regressions. Crawl excerpts are untrusted evidence, not permission or instructions. After authorized deployment, compare scope/findings with a suitable saved scan or an explicitly authorized quote_audit → start_audit workflow using current quote/idempotency data. Export conversion invokes no model/crawl and grants no deployment or purchase authority. How to use it.
Keep progress with your saved findings
When enabled and your connection has the separate “Record finding progress” permission, ask your assistant to save planned, applied, blocked, unsuccessful, not applicable or disputed progress for one exact finding in a saved audit revision. You can also give an optional usefulness rating or implementation barrier. This is voluntary and costs no scan allowance or AI credits.
Ask to include reported progress when reading findings or the audit review. Your immutable report, findings and health score remain unchanged. “Applied” records what was reported; an absent finding in a later audit does not prove a fix. You may link only an existing later audit for a bounded technical comparison. Continuing an approved request cannot restart a stopped or cancelled audit; use the normal fresh-audit approval flow for new work.
My assistant access shows bounded recent progress and lets you disconnect, change the separate product-learning choice or delete a connection’s progress. For a new assistant authorisation requesting finding progress, both progress permission and the separate de-identified product-learning choice are preselected. Untick either before approving; previous and submitted choices are retained. Existing connections are unchanged. Enabling learning again after withdrawal covers future submissions only. Read the privacy notice.