Built an MCP server so Claude can query my analytics directly. Use cases I did not expect
Last week I shipped a Model Context Protocol server for my analytics SaaS. Now Claude (or Cursor, or any MCP compatible client) can directly query traffic, revenue, and funnel data.
I expected this to be a nerdy feature for power users. Three surprising things happened:
-
Customers started asking Claude weekly questions like 'what changed in my traffic last week that explains the revenue drop'. Claude pulls the data, finds the regression, writes a summary. Saves them a dashboard deep dive.
-
One customer built an internal slack bot in 20 minutes that runs
/traffic-yesterdayand posts a summary to their team channel. Just claude desktop + the MCP server, no custom code. -
I personally use it to write my own weekly investor update. 'Pull last week's metrics, compare to previous week, give me a 3 paragraph summary in the tone of a casual update'. Used to take me 90 minutes. Now 5.
The MCP integration changed analytics from 'open dashboard, scan numbers, write report' to 'ask, get answer'. For most people that is the actual workflow they want.
If you want to play with it: zenovay.com, it ships on the paid plans.
Any other MCP server builders here? Curious what use cases you have seen that surprised you.
Valerio
Congrats on shipping the MCP server for your analytics SaaS — Claude querying traffic/revenue/funnel data directly is a strong use case. Since it's brand new, worth locking down scoping now before it gets more usage (read-only where possible, revenue data flagged as sensitive in any logging). I do exactly this as a service — happy to do a quick free pass if useful.
The Slack bot use case doesn't surprise me, it's the natural next step once the data stops being locked behind a dashboard. I've been running GA4 and Search Console the same way, plain-English queries against the live account instead of exporting anything, and the thing that actually changed my week wasn't any single query, it was being able to hold GSC data, GA4 and newsletter numbers in the same session and reason across them. "The traffic spike is from one Reddit thread, the query movement is unrelated, and the subscriber bump came from Tuesday's article" is a sentence no single platform's UI produces, because no single platform's UI holds all three datasets at once.
The failure mode worth watching for: given a table of numbers, the model will offer a confident explanation for any movement, and it's sometimes wrong. Caught one recently where it explained a spend anomaly as creative fatigue when the real cause was a budget change two weeks earlier. Worth keeping a human in the loop on the "why," not just the "what."
I wrote up the wider setup here: dhawalshah [.] net/article/claude-code-for-marketing/