The MCP gold rush is real — and nobody's selling the shovels

3 new cards today, 125 signals scraped, and producthunt keeps delivering the same pattern: devs shipping tools, then realizing they're flying blind.

The one that actually made me stop scrolling: Analytics and evals for MCP agent sessions. 78分. This is the classic "I built the thing, now I need to know if it works" pain — and it's hitting everyone who jumped on the MCP bandwagon. You've got thousands of devs building MCP servers this year, and most of them have zero visibility into what happens after they ship. Session logs, failure rates, token usage, eval suites — that's the kind of tooling that becomes sticky because it's not a feature, it's infrastructure.

Monetization? Subscription based on session volume is smart. You're not selling a tool, you're selling the data layer. Start with a free tier for personal projects, charge once teams need to share evals across their org. I'd build this with a simple stack — Postgres for storage, a lightweight ingestion API, and a Next.js dashboard. Don't overthink it.

Now the bookmark stuff. Bookmarks saved but never found at 70分 and Video bookmarking and timestamp manager at 65分. Look, I've seen a dozen of these. The problem isn't that people don't need better bookmarking — it's that they don't want to maintain another system. You're competing with "just Google it again" which is free and instant. The video timestamp thing is slightly more interesting because it's solving a real workflow pain for people who learn from tutorials, but the freemium model with cloud sync means you're in the storage business, and that's a race to the bottom.

Honestly? The bookmark ideas are fine side projects. You might make a few hundred MRR if you nail the AI search angle. But the MCP analytics play has actual legs because it's B2B, it's recurring, and the market is growing fast. Every MCP server that launches tomorrow is a potential customer.

If I were building today, I'd take the MCP analytics idea and go deep. The devs who need this are already on Twitter/X and Discord, they're frustrated, and they'll pay $20–50/month to stop guessing. I'd have a landing page up this weekend, scrape a few hundred MCP servers from GitHub, and cold DM the maintainers.

That's where I'd put my time. The other two are solutions looking for a problem that's already solved by laziness.