Guided onboarding and playbooks over MCP

You do not have to know what SPF is, what an exposure scan does, or which of nine setup steps applies. Over MCP, adding a domain is a conversation, and the server carries the playbooks.

Adding a domain

The assistant calls plan_domain_onboarding (optionally with the domain). It returns the questions to ask, in order, with why each matters: does the domain send email; which add-time checks to run (lookalikes, subdomains, vulnerabilities); the accessibility audit and its 5-credit cost; whether to keep scanning for exposure; whether it is a local business (name, city, phrases); where alerts should go; an address at the domain for the ownership link; which mailboxes to watch in breach data; which group or client it belongs to.
The assistant asks in plain language, one or two questions at a time.
onboard_domain applies every answer in one write. Nothing in it spends credits; the paid steps (Local SEO quick start, the accessibility audit) are quoted separately.
Every step reports (done, skipped, refused with a sentence), and the response includes the setup guide so the assistant can say what is done and what is left, in your words. It never ticks a step itself.

In Claude Code the whole thing is /mcp__domainguard__onboard_domain. add_domain still works for a bare add.

Setting up Local SEO

get_local_seo_quickstart says what is saved and the exact price of a first measurement. run_local_seo_quickstart saves the business and up to five phrases, confirms the city is covered, then checks the map pack and the organic result for every phrase and reads the Google Business Profile: two credits a phrase plus two. The assistant quotes that number and waits for a yes. Prompt: /mcp__domainguard__set_up_local_seo.

The playbooks

Prompt	What the assistant does
`review_domain`	Reads everything stored about one domain, verifies what it can (resolves the DMARC record and compares it with what was quoted, fetches the page and reads the headers, counts the scripts), reports any disagreement, and delivers the fixes in order
`portfolio_triage`	One ranked list across every domain, from stored data, spending nothing
`fix_email_deliverability`	SPF, then DKIM, then DMARC, then the report stream, using the setup checklist and the ranked diagnostic

How the assistant is told to behave

The operating guide at nhmohio.com/mcp/claude-guide.md sets the rules: prefer get_checkup_history, get_seo_results and get_vuln_scan_latest over new scans; call get_local_seo_catalog before running a catalog action; use a fresh request_key for every paid run and get_local_seo_operation to recover one after a timeout; treat an unconfirmed verdict as "not looked", never as safe; and say what the scanner said and what it found when they differ.

Common questions

Can Claude change my DNS? No. Nothing in the API touches DNS. It can tell you the record and you paste it.

Can it spend without asking? Only if you gave it a write key and told it to. The guide instructs it to quote and wait; the quote endpoints exist so it can.