Metrifact connects Claude to your Google Ads, Meta Ads, GA4, Search Console, and Shopify accounts. Connect once, then ask in plain language. Every answer reads your live account data, not a cached report. Fourteen skills cover the actual jobs a performance marketer or agency does every week: a morning health check, a growth plan, a full account audit, a client report. You don't need to know their names. Say what you need, and Metrifact routes you to the right one.
Checked whether the spike was even real before anything else. The account only ran ads on one day the prior week versus four this week. Per day, spend never changed.
InvestigateFound $340 in wasted search terms and one stalled ad set. Ranked by dollar impact, not by count - the ad set matters more even though there were more search terms.
OptimizeFrequency is up, but CTR hasn't moved. That's rising auction competition, not fatigue - no refresh needed yet.
Creative RefreshChecked tracking before performance. This campaign's conversions aren't being counted right now - that's the finding, not "cut it."
Scale or CutOne file. Installs all 14 skills and the Metrifact connector together. How it installs
"anything on fire" · "morning check" · "how are my accounts" · "did anything break"
Checks broken tracking, disapproved ads, delivery that stopped, spend anomalies, pacing drift, and out-of-stock products still being advertised, across one client or every client.
A quiet day gets a quiet answer. It won't invent a problem to justify the time it took.
"how are all my clients doing" · "how's the book looking" · "what should I look at first today"
Reuses Daily Check's own logic across every connected client, then ranks by what needs attention first.
Broken measurement jumps to the top of the list even when everything else looks fine, because a number you can't trust is worse than a number you don't like.
"how should I plan spend" · "what should I launch next" · "how do I split budget between Google and Meta"
Builds from real organic search demand, site behavior, and sales data when there's no ad history yet, or from the account's own spend-to-result curve when there is. Compares Google and Meta directly when both are connected.
It won't compare Google against Meta until it's confirmed both platforms' tracking can be trusted. A reallocation built on a broken number is worse than no reallocation at all.
"launch a new campaign" · "build out this account" · "turn that growth plan into a real campaign"
Handles structure, keywords or audiences, a tracking check before building anything, UTMs, and a real pre-launch QA pass.
Nothing goes live on its own. Every campaign it builds sits paused, waiting for you to turn it on.
"write new ad copy" · "give me headline variants" · "write creative for this product"
Checks whether a real top performer actually exists in this account's own data before writing anything.
If the account doesn't have enough history to prove a pattern works, it says so instead of dressing up a guess as a proven winner.
"optimize this account" · "find savings" · "what should I change this week"
Finds wasted spend everywhere it usually hides: search terms, bids, budgets, audiences, placements, and stale creative.
Small findings don't make the list just because they're real. A $15 fix riding along on a report about thousands of dollars gets cut, not padded in for volume.
"why did my CPA spike" · "why did conversions stop" · "why did traffic drop" · "why is CTR collapsing"
Confirms the symptom actually happened, at the right grain, before doing anything else, then branches by what the account's own goal is.
A CPA spike is often the account running fewer days that week, not a real cost increase. Divide by active days before blaming the campaign.
"audit this account" · "new client, what am I working with" · "is my SEO healthy"
Covers structure, tracking, historical trajectory across all available history, current waste, governance risk, and SEO health when Search Console is connected.
It leads with the account's full trajectory, not this week's snapshot — the view a one-off audit can give that a recurring report never has time for.
"is my creative fatigued" · "why is my CTR dropping"
Looks for a real trend, not a single snapshot, sustained over one to two weeks rather than one bad day.
Rising frequency alone isn't fatigue. CTR has to be falling at the same time, or it calls that something else, usually more competition in the auction, and says so instead.
Name a competitor and ask how their page stacks up.
Pulls the client's own real ad copy and landing page first, then the competitor's, compared in both directions rather than described alone.
The client's own page usually turns up the bigger problem, not the competitor's.
"is my ROAS real" · "what's my true CAC" · "do my numbers add up"
Reconciles platform-claimed results against a neutral source (GA4 or Shopify), checks for cross-platform double-counting, and computes blended efficiency across every connected platform.
If it can't verify a number, it won't hand you one. When a connection has expired, "fix that first" is the entire answer, not a false under-counting finding that would look identical to a real one.
"should I kill this campaign" · "is this channel worth keeping" · "should I scale this up"
Checks tracking integrity and how much real evidence actually exists, before looking at performance at all.
Zero conversions on broken tracking isn't proof a campaign failed. It's proof nothing was measured, and that gets fixed before anything gets cut.
"help me set up an A/B test" · "is my test done" · "what does my test say" · "should I promote this experiment"
Handles two separate jobs, kept separate: sizing a new test to what the account can actually support, or reading an existing test's real progress.
Waiting on a result doesn't make an unfinished test finished. It tells you how much longer, not a verdict it can't back yet.
"write a client report" · "weekly update" · "monthly report" · "exec summary"
Pulls the numbers once, at the right grain, then works with any platform combination the client actually has: Google Ads alone, Meta alone, either paired with GA4, or the full stack.
A number moving isn't the story. When spend drops to zero on three campaigns because someone consolidated budget on purpose, that's one sentence, not an alarm.
Download the plugin
One file. It contains all 14 skills, the router that picks between them, and the Metrifact MCP connector.
claude.ai → Settings → Customize → Plugins → Upload a plugin
Select the file you downloaded. claude.ai's own upload screen accepts a
.zip
or .plugin archive containing a
.claude-plugin/plugin.json manifest.
Authorize the Metrifact connector, if prompted
One install registers both the skills and the connector together. If you've never connected Metrifact before, claude.ai may prompt you to authorize it, the same OAuth step as connecting any account, at this point or the first time a skill actually tries to read your data.
Start a new chat
Skills mount when a conversation starts, not mid-conversation. If a chat was already open when you installed, they won't appear in it — start a new one.
Ask in plain language
You don't need to open a specific skill. Ask what you actually want to know, and Metrifact routes to the right one.
Two things worth knowing
Skills are per-surface. If you use both claude.ai and Claude Code, you'll need to add them in each place separately - installing the plugin in one doesn't carry over to the other.
Uploaded plugins aren't controlled by or verified by Anthropic. claude.ai links out to each plugin's own homepage for exactly this reason - this page is that homepage for the Metrifact plugin.
Just want one skill?
Every skill is also available as a single .md file if you don't want the
full plugin. Browse the individual files.
The plugin is the primary install; these are the fallback.
A router skill reads what you actually asked and picks which of the 14 skills above answers it, then tells you which one it used. "Why did my CPA jump" and "anything on fire" go to different skills without you ever needing to know that Investigate isn't Daily Check. If nothing is connected yet, it says so and points you at setup instead of guessing.
No. Custom skill and plugin uploads work on Free, verified directly in a Free account, along with every paid plan.
No. Every skill reads your live Google Ads, Meta Ads, GA4, Search Console, or Shopify data through the Metrifact connector. There's no version of this that works from a pasted report or without a live connection. The plugin installs the connector alongside the skills, so this isn't a separate step.
Metrifact builds and maintains all 14, plus the router that picks between them. They're versioned and available as a single plugin download, or as individual files if you'd rather add one at a time.
Yes. Every skill is also available as its own .md file. The plugin is the
faster path if you want the whole set plus the connector in one step.
Ask in plain language. A router skill included in the plugin reads the request and picks the right one for you, then tells you which skill it used.
Download the skills plugin above, or connect Claude to Metrifact directly and explore the tools yourself.
Start free trial