Privacy + safety model

Agent-WebmasterID is a free SEO intelligence + AI-assisted operations layer. The agent reads verified Google Search Console data and the operator’s own watchlist, runs deterministic analytics on that data, and produces Claude-ready prompts the operator copies into a tool of their choice. Every workflow requires explicit operator approval before it runs.

What the agent never does

No SERP scraping. The agent only calls the official Google Search Console API.
No invented data. Volume, difficulty, and SERP positions beyond GSC’s average position are never produced.
No guaranteed rankings. Recommendations are advisory.
No automatic publishing. Output is a prompt, checklist, or markdown brief.
No auto-deploy. The agent does not push code, open PRs, or modify your live site.
No DNS or billing changes. The agent has no integration that could alter those.
No background MCP write tools. The MCP bridge is read-only.

Verified vs manual data

Every keyword in the workspace carries a dataSource label:

Verified — the row is populated from a recent GSC fetch. Clicks, impressions, CTR, and average position come straight from the official Search Console API.
Manual — the operator typed the keyword themselves. The agent surfaces the operator-typed currentPosition with a “(manual)” suffix so it’s never blended with verified GSC averages.

Operator approval is mandatory

Every workflow follows Analyze → Prepare → Review → Approve → Execute. The agent’s execute function rejects any workflow that isn’t in the approved state. Workflow outputs are limited to prompts, checklists, and markdown briefs the operator copies into a tool of their choice.

Tokens + secrets

Google OAuth refresh tokens are encrypted server-side (the M17 vault) and never returned to the browser.
MCP tokens are stored as SHA-256 hashes. The plaintext is shown once at issuance and never echoed again.
Workflow plans + results carry only operator-readable scalars and the safe Claude prompt. Raw API responses, OAuth payloads, billing secrets, and database URLs are never written.

Data scope

The agent only reads from the workspace’s own Search Console properties.
The agent does not pull from competitors, third-party trackers, or any unofficial source.
Workspace data is RLS-gated server-side; cross-workspace ids are silently invisible.