Yunus Emre Vurgun
I can see myself using an agent to find the issue and then opening the dashboard to double-check what it found.
The paid in Stripe, still expired in Firebase bug is a good test case, because it genuinely needs four tools and someone copying IDs between them. I'm glad you still read every line of the fix PR. Tool count has been my biggest headache with MCP setups, so the 77,000 to 8,700 token drop is the number I'd want to reproduce. Which gateway handled that bug for you?
The tool-search gateway point matches what I've seen: the win isn't just fewer tokens, it's fewer wrong tool picks when two servers expose near-identical tools (e.g. three different 'search' tools). One thing I'd keep a dashboard for, though, is the stuff nobody thinks to ask about: an agent answers the question you asked, while a well-designed dashboard surfaces the anomaly you didn't know to ask about. A pattern that works nicely is having the agent run a small set of scheduled 'standing questions' and only alert on deltas, so you get the proactive part without staring at charts. Also +1 on evals for skills; a handful of golden queries per data source catches most regressions when the model changes.