Integrations · MCP
Your ad performance, inside Claude.
Sturnix ships a Model Context Protocol server, so Claude can read your live spend, pull the waste audit, and prep actions — every call through official APIs, every change inside your guardrails.
Setup
Three steps to connect.
Connect once
Add the Sturnix MCP server in Claude. Scoped, read-first, revocable any time.
Claude reads the truth
Live spend, the five-dimension waste audit, and channel splits — straight from Sturnix.
You stay in control
Claude preps actions into your review queue. Guardrails hold; nothing executes unapproved.
What you can ask
Six ways in, one source of truth.
"Where am I wasting the most?"
The live five-dimension audit, in dollars, with the specific campaigns and ad sets behind each number.
"Why did CPA jump on Tuesday?"
Signal-level answers from your own data — frequency, CPM repricing, creative fatigue — not generic guesses.
"Queue the fix."
Claude preps the action into your Sturnix review queue with reasoning attached. You approve; the engine executes.
"Meta vs Google this month?"
Cross-platform splits no single platform will show you, straight from the engine.
"Am I near any caps?"
Live guardrail status — spend caps, floors, no-touch zones — before you plan the next move.
"Draft the Monday summary."
Reallocations, ROAS movement, and actions taken — pulled from the ledger, ready to paste.
The security model
Read-first. Execution stays inside Sturnix.
The MCP server is read-first: Claude can query spend, audits, and channel splits, but the only write it can perform is placing a proposed action into your review queue. Nothing executes from a conversation. Execution stays inside Sturnix, behind your autonomy mode and your guardrails, through official platform APIs — logged and reversible like every other action.
Access is scoped and revocable: disconnect the server and it's gone. Your ad-platform credentials never pass through the model; Sturnix holds the platform connections, and Claude only ever talks to Sturnix.
MCP FAQ
Answered directly.
The Model Context Protocol — an open standard that lets AI assistants like Claude connect to external tools and data sources. Sturnix ships an MCP server, so any MCP-capable client can read your ad performance and queue actions through the engine.
No. Claude can propose; only the Sturnix engine executes, and only inside your autonomy mode and guardrails. A proposal from a conversation gets exactly the same treatment as one from the engine itself: queued, reviewed, logged, reversible.
Anything that speaks MCP — Claude Desktop and Claude Code today, and the standard is spreading fast. If your tool of choice supports MCP servers, it supports Sturnix.
Get started
Bring your spend into Claude.
Connect once. Claude reads the truth and preps the fix — you stay in control.