Prompts
ThriveStack ships native MCP prompts — system-prompt-style personas and setup wizards you invoke directly from your AI tool’s prompt picker.
Prompts are a built-in MCP primitive. Instead of describing how you want the assistant to behave every time, you invoke a single named prompt and it takes on that persona or runs that workflow for the rest of the conversation.
Prompts are versioned with the ThriveStack MCP server. When we ship a new one, it appears in your client automatically. No install.
None of the current prompts take arguments — invoke them as-is, no project or brand selection needed up front.
How to invoke a prompt
The command surface depends on your client.
| Client | How to invoke |
|---|---|
| Claude (Desktop, Web, Code) | Type / in the message input and pick the prompt from the list |
| Cursor | Type @ in the chat and pick the prompt |
| Other MCP clients | Check your client’s documentation for prompt support |
If prompts don’t show up after connecting, remove and re-add the ThriveStack integration to refresh the tool and prompt list.
Available prompts
ai_visibility_analyst is the flagship prompt today, matching the AI-Visibility-focused tool rollout (see Tools). Two setup wizards are also available for instrumenting new marketing sites or products.
ai_visibility_analyst
AI Visibility content strategist. Reads citation-gap probe evidence (get_citation_probe_data, get_domain_report) and source-domain metrics, then works through a structured process:
- Pulls the domain’s latest visibility report and raw citation-gap evidence
- Classifies each gap into a tier — Silent (never mentioned), Mentioned-not-cited (brand comes up but no source is cited), or Losing share (cited less often than competitors)
- Pulls owned and earned content recommendations for the gaps that matter most
- Optionally drafts a content outline for the top recommendation via
generate_content_outline
No arguments.
Use when: you want a structured pass at “why aren’t AI engines citing us for this topic, and what should we publish to fix it”, rather than pulling the raw probe data and recommendations yourself one tool at a time.
“Run the AI visibility analyst for ourdomain.com”
Follow up with a specific competitor, topic, or content type to drill in — the assistant still has every AI Visibility tool available after the prompt runs.
thrivestack_marketing_setup_wizard
Marketing telemetry installation wizard. Returns a wizard prompt meant to be pasted into Claude Code inside your marketing website repo, walking it through instrumenting the site with ThriveStack visitor tracking.
No arguments.
Use when: you’re setting up ThriveStack tracking on a marketing site for the first time.
thrivestack_product_setup_wizard
Product telemetry installation wizard. Returns a wizard prompt meant to be pasted into Claude Code inside your application repo, walking it through instrumenting your web app with ThriveStack product telemetry.
No arguments.
Use when: you’re setting up ThriveStack tracking on your product for the first time.
Combining prompts with free-form questions
Prompts are a starting point. After invoking one, follow up with questions in plain language — your AI assistant still has access to every tool, so you can drill into any number, account, or domain the prompt surfaced.
“Run the AI visibility analyst for ourdomain.com, then show me the raw citation evidence for the gaps it flagged”
“Now check the same gaps for [competitor]’s domain”
Every currently-live ThriveStack MCP tool and prompt is read-only — they only read the same ThriveStack data you can already see in the dashboard.