Orphex MCP - Run your entire digital marketing stack via one unified MCP

@yagiz_karakaya Yes. Orphex has built-in Impact Analysis module which runs a Causal Impact model: it builds a counterfactual of what would have happened without your change, then reports the incremental lift with a 95% credible interval and a significance verdict for each segment, so you can tell a real effect from noise.

You don’t need to prepare any data. Orphex uses the data from your connected platforms and handles the time series, the pre- and post-period windows and the control signals itself. It even scans your accounts for recent changes, like a budget shift, a new campaign or a bid strategy switch, and suggests which ones can actually be measured. You can just ask your assistant something like “Did last month’s budget increase on our Meta prospecting campaign actually drive incremental sales?”

@alpkosar Great question. Orphex doesn’t add its own tracking pixel. It connects directly to where the sale actually happens and reads each source’s own attribution side by side:

• Your store: On Shopify, for example, orders and revenue are tied back to campaigns, UTMs and referring channels under first-click, last-click, last non-direct click and linear models, so you can see how the credit shifts with the model.

• Your analytics and MMP: GA4 traffic sources and key events, and Adjust, AppsFlyer or Singular for apps.

• Each ad platform’s own view: what Meta, Google or TikTok claims, under its own attribution window.

Orphex doesn’t blend these into a single “truth” number, because they measure different things. It shows where they agree and explains why they differ. For the deeper question, “would this sale have happened without the ad?”, Impact Analysis or MMM module measures the incremental effect causally.

@hakan_cadirci Thanks! Yes, with one nuance. Every proposal comes back as a structured preview before anything runs: the target’s current value read live from the platform, the exact values that will be sent, the guardrail verdict, and the account and currency it applies to. The proposal itself is plain JSON in the tool call, so developers see exactly what was asked for.

The confirm token is cryptographically bound to that preview. What executes is exactly what you approved, and if anything changed in between, the confirm is refused. For some platforms the preview also includes the native request body, and where the platform supports it (Google Ads, for example), the change is validated by the platform itself before you confirm. For other platforms you see the normalized fields rather than the raw HTTP request.

There’s also a readiness check for every action: it tells you whether a call would be accepted, and why not, without creating a proposal at all.

@oleksii_sekundant Exactly, that’s the gap we kept hitting ourselves. All advertising platforms can both “own” the same conversion, because each one counts it under its own attribution method and window. Orphex shows every platform’s number side by side, each tied to the platform that reported it, and never quietly merges them into one total. You can also switch attribution windows, for example 1-day vs 7-day click, and watch how each platform’s count shifts. When the numbers disagree, you can see exactly why. Thanks for calling it out!

@ebru_sevik Thank you! All three, set as workspace policy rather than per prompt, so they hold no matter what the assistant is asked:

• Preview first. Every change comes back as a preview showing the before and after values. Nothing goes live until it’s confirmed.

• Human approval. You can require a recorded approval for every change, or only when a change is bigger than a % threshold you set. You can also require a second approver and limit which roles can approve.

• Spend brakes. You can set a max % change per action/prompt, a daily cap on budget increases, a monthly spend envelope with an early warning, daily action quotas, and a budget ceiling for anything. You can also lock campaigns by name so they’re never touched.

Anything over a limit is refused before it reaches the ad platform, and the assistant tells you which rule blocked it.

@thamibenjelloun yes you can define pre-approved change limits as you wish!

@berrinkarakas Thank you! Orphex MCP doesn’t store your performance or spend data, neither on your machine nor in our cloud. When you ask a question, Orphex fetches the numbers live from the ad platform through its official API, hands them to your AI assistant for that answer, and keeps no copy. Every answer reflects what the platform reports at that moment.

Access goes through each platform’s official OAuth, scoped to your workspace, and you can revoke it anytime from the platform’s own settings.

@fatih_eren_saribuga Great question. That’s exactly the trap Orphex is built to avoid. Its diagnosis method separates measurement from performance before anything is recommended:

1. Split the drop. Spend, conversions per spend and value per conversion, then the funnel: CTR, CPC and conversion rate. Conversions collapsing while impressions and clicks hold steady is treated as a measurement break until shown otherwise, not as an ad problem.

2. Check tracking at the source. For example, each Google conversion action’s status and when it last received data, Meta’s tracking specs on the ad, and your GTM container (paused tags, consent settings, live version vs. draft). A “tracking is broken” verdict has to rest on that evidence, never on a falling count alone.

3. Compare against the store. If the platform’s conversions fall but your Shopify orders don’t, it’s attribution or tracking. If orders fall too, while clicks and traffic still arrive, look at the site or checkout.

4. Rule out the quieter causes. A changed attribution window or conversion value setting, conversions still arriving for recent days, edits made in the platform’s own UI, or ads stuck in review.

Only when all of that is clean does it call the drop a performance issue. Even then, it starts with a read-only diagnosis; any budget or bid change goes through preview and your approval.