Building Custom MCP Tools with Sentrum API + MCP Access.

Pro's MCP server exposes Sentrum's default diagnostic tools — campaign diagnostics, creative analysis, account context, recommendations, askQuestion, and more. For most people that's the whole workflow. But once you've been using Sentrum from Claude Code or Cursor for a few weeks, patterns emerge that the defaults don't cover.

Custom tools solve that — and they live on your side of the wire. You define a tool in your agent's own toolkit (a Claude Code skill, a Cursor rule, or your own MCP server), and its handler calls Sentrum's MCP tools and API underneath — same auth, same workspace scoping, same Sentrum account context.

What a custom tool looks like

A tool is three things: a name, an input schema, and a handler that returns JSON. Example: an agency that wants a "client-ready Slack message" tool that pulls the week's diagnosis from Sentrum and formats it in their voice:

// Define the tool in YOUR agent's toolkit / MCP server —
// its handler calls Sentrum's MCP tools underneath
{
  "name": "myAgency.draftClientSlackUpdate",
  "description": "Drafts a 4-sentence Slack update for the agency
    client in our house voice. Always opens with the headline number,
    closes with the next action. No jargon, no 'the algorithm'.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "clientWorkspaceSlug": { "type": "string" },
      "dateRange": { "type": "string", "enum": ["7d", "14d"] }
    },
    "required": ["clientWorkspaceSlug"]
  }
}

The handler runs wherever your agent runs. It can call Sentrum's MCP tools, your own database, or external APIs you've authorized. The agent sees one tool call; under the hood it can fan out to whatever you need.

Three patterns this enables

Pattern 1 — House-voice client reports

Agencies have a voice. Default LLM output doesn't. Wrap your tone, your closing CTA, your "always include profit-ROAS" rule into a custom tool. Your agent reaches for it when you ask for a client update, the output ships in your voice every time.

Pattern 2 — Custom scoring

Sentrum's default diagnostics use a generic-but-good scoring system. If you've developed your own scoring — say, a creative-fatigue index weighted by your specific funnel data — you can wrap it as a tool. The agent calls myCo.creativeFatigueIndex and gets your score back, computed over the data Sentrum returns.

Pattern 3 — Cross-source merges

The Meta API knows Meta. Sentrum knows Meta + your account context. But neither knows your CRM. Wrap a tool that joins Sentrum data with your customer LTV table, and the agent can answer "which creative drove highest-LTV customers" — a question neither system can answer alone.

Webhook subscriptions (Agency plan)

Custom tools are pull-style — your agent has to ask. Webhooks are push-style — Sentrum tells you when something happens. They're part of the Agency plan. Available subscriptions:

• insight_generated — fires when a new insight completes
• alert_triggered — fires when an alert condition is met
• recommendation_created — fires when a new action recommendation is generated
• sync_completed — fires when an account data sync finishes

Each fires a signed POST (HMAC-SHA256) to your endpoint with a typed payload. Wire to Discord, Slack, your ops dashboard, or directly into another MCP tool to trigger downstream automation.

What the allowance looks like

Each plan carries a monthly AI allowance — the current figures per plan are on the pricing page. MCP requests pass through the same rate limiting as the rest of Sentrum's AI endpoints, which exists to catch runaway scripts and leaked keys, not to get in the way of a working agent loop.

If you're running an agent loop heavy enough to hit the ceiling, talk to us — the Agency plan carries the highest monthly AI allowance.

What's included on Pro

The full Soren diagnostic engine (multi-pass diagnostics, account memory, anomaly detection, creative analysis), the Creative Intelligence suite, campaign wrap reports, PDF exports — plus the API keys and MCP server access this post is about, with every default diagnostic tool. Your custom tools layer on top.

Who this is not for

If you're using Sentrum from the dashboard, you don't need any of this. The dashboard covers the whole diagnostic workflow. API + MCP access is for the workflow shift — when you're running media buying through an AI agent more than through the dashboard, and you want the agent working from Sentrum's diagnostic data instead of raw CSVs.

Most users start in the dashboard and reach for the MCP server once they've built a few prompt patterns they want to wrap into tools.