Skip to content
Docs
Tutorial2 · Set up2.4 · What to do with MCP today
For workspace members

2.4 · What to do with MCP today

The tasks worth stealing first, once your agent can interrogate the workspace in plain language.

Open in ChatGPTOpen in Claude

Once your agent is connected, your workspace, the container holding everything GrowthOS knows and does for your brand, becomes something you can interrogate in plain language. The tasks below are the ones worth stealing first: each pairs a real job with a prompt you can paste as-is (swap in your own pages and workspace name). They're ordered by how often teams reach for them.

1. Diagnose an underperforming page

The everyday one. GrowthOS packages the handoff: any page in your Pages Portfolio (the list of every page on your site) has an Analyze action that builds the full diagnostic prompt for your agent: opened straight into Claude on the web, or copied to your clipboard to paste into Claude Code or Cursor. Or ask directly:

Using GrowthOS, diagnose /blog/example-page. Pull its health and quality audits, traffic trend, and keyword rankings. Tell me the single biggest bottleneck and what you'd change first.

The agent reads the audits, the trend, and the striking-distance keywords, and argues a fix instead of handing you a dashboard.

2. Publish a ready brief

Covered in the shipping guide: the Publish step's agent card hands your agent a prompt that reads the finished brief, the plan for one page, from GrowthOS and drafts it into your CMS through your CMS's own MCP. Drafts only, never live. Use the built-in card rather than improvising the prompt; it carries the full handoff instructions.

3. The Monday portfolio check-in

Using GrowthOS, give me a portfolio check-in: overall summary, traffic by channel versus the previous period, and any pages whose momentum turned Declining or Freefall. End with the three actions you'd take this week.

Declining and Freefall are 2 of the 5 performance bands, the labels describing the direction of a page's traffic. 5 minutes, and your week starts with a triage list instead of a feeling.

4. Scale a winning page

When one page earns real traffic, there's usually a pattern worth repeating:

Using GrowthOS, look at /blog/winning-page. Work out why it wins structurally (format, intent, coverage). Then propose five sibling pages that apply the same pattern, and check each against my existing pages and opportunities so nothing cannibalizes.

Take the survivors to the opportunity finder as seeds. An opportunity is a proposed page worth writing or rewriting.

Using GrowthOS, my new page /blog/new-post is about to go live. Find the best internal link candidates: existing pages that should link to it, and pages it should link to. Give me anchor text suggestions for each.

Internal linking is the boring task with the biggest payoff in content, which makes it well suited to agent work.

6. Read your AI traffic

Using GrowthOS, how much of my traffic comes from AI assistants (ChatGPT, Claude, Perplexity, Gemini), which pages do they land on, and how does my AI visibility look across my tracked segments?

Segments here are your AI visibility segments, the slices of your market GrowthOS watches inside AI answer engines. This is the early-warning surface for the AI side of your bet: assistants often cite pages before rankings move.

7. Audit a draft's sourcing

Using GrowthOS, pull the research behind the brief titled "Example article". Check the draft's main claims against that research and flag anything that isn't supported or needs a source.

The research behind every draft is readable over MCP, so claim-checking becomes a delegable job rather than a scavenger hunt.

Writing your own

Everything above composes from the same read-only toolkit: pages and their audits, traffic and keywords, briefs and their research, opportunities, and your context documents, what GrowthOS knows about your company. Two habits make custom prompts land: name your workspace when you have more than one, and refer to pages by URL path. Ask for a recommendation, not only data, since the agent has the context to argue.

Last updated at August 13, 2026

On this page