Diagnose Meta Campaigns from Claude Code in 5 Minutes.

If you live in Claude Code, Cursor, or Claude Desktop, you can read your Meta ad account from inside the chat. No new dashboard, no copy-pasting numbers. Sentrum's MCP server exposes five tools your AI agent can call directly. Each tool returns the same diagnostic shape you'd see in the dashboard.

Here is how to set it up and what to ask first.

Step 1 — Generate an API key

In Sentrum, go to Settings → API Keys. Click Generate key. Copy the string that starts with skx_live_. Keep it somewhere safe — you only see the full key once.

API keys are workspace-scoped. If you run an agency, give each client their own key.

Step 2 — Wire it into Claude Code

One command:

claude mcp add sentrum https://sentrum.app/api/mcp \
  --header "Authorization: Bearer skx_live_your_key_here"

Restart Claude Code. Type /mcp and confirm sentrum is listed with five tools.

For Claude Desktop and Cursor, the config is JSON instead of a command. See the prompting guide for the snippets.

Step 3 — Ask three diagnostic questions

The five MCP tools are:

• sentrum.getCampaignDiagnostics — campaign KPIs + anomalies + multi-pass diagnosis
• sentrum.getCreativeAnalysis — top creatives by spend with fatigue + vision tags
• sentrum.getAccountContext — account memory: baselines, patterns, recent anomalies
• sentrum.getRecommendations — pending actions with severity + Ads Manager deeplinks
• sentrum.askQuestion — natural-language Q&A with full account context

You do not have to memorize them. Just ask in plain English and Claude reaches for the right one. Three starters that produce useful answers on day one:

1. The Monday morning check

What changed in my Meta account this week? What should I do about it?

Claude calls getCampaignDiagnostics with a 7-day window and walks you through what moved, why, and the highest-leverage action. Same logic that runs in the Daily Pulse tab.

2. The creative read

Pull my top 10 creatives by spend. Flag fatigue. Tell me what's
working in the winners — hook type, message theme, visual style.

You get a ranked list, fatigue flags, and a pattern read across the winners. Useful right before a creative brief.

3. The reallocation question

If I shift 30% of spend from my worst-performing campaign to my
best-performing one, what does Sentrum think happens to blended CPA?
Reason against my account baselines.

This is where the MCP layer earns its keep. The raw Meta API can't answer hypotheticals. Sentrum's account context — your baselines for CPM, CTR, CVR — flows into Claude's reasoning so the answer is grounded in your account, not in generic ad-platform knowledge.

What you should not do

Sentrum's MCP is read-only. You can't change budgets, pause ad sets, or upload creatives through it. That is deliberate. Your ad account can't be edited by accident. If you want to act on a recommendation, the response includes an Ads Manager deeplink and you make the change yourself.

Also — don't use the Sentrum MCP for raw data retrieval. If you just want a list of campaign IDs, use Meta's own CLI. Sentrum is for the layer above raw data: the diagnosis, the comparison, the action.

Where this goes next

Most users settle into a daily pattern: one Monday-morning question, two or three creative reads per week, and ad-hoc reallocation questions when something moves. The dashboard is still there when you want to stare at charts. The MCP layer is for when you don't.

For the full prompt patterns, workflows, and the Developer tier (2,000 AI calls/day, custom MCP tools, webhook subscriptions), see the prompting guide and the pricing page.