MCP hit 97 million installs in March 2026.
That's not a developer milestone. That's a distribution signal.
Here's what's actually happening: the AI tools your customers use every day — Claude, Copilot, Cursor, Gemini, and a dozen others — now connect to external applications through Model Context Protocol. When a user tells their AI assistant to "check my calendar and cancel tomorrow's appointment," the agent can only do it if the scheduling software has an MCP server.
Otherwise, it can't see the product at all.
Doesn't recognize it. Doesn't know it exists. Can't bring it into the workflow.
This is the same story as 2012. SaaS products without REST APIs got left out of the integration ecosystem. No Zapier. No partner programs. No "works with X" badges. Their product wasn't bad — it just couldn't be reached.
The same thing is happening now with AI agents. Only faster.
When Anthropic launched MCP in November 2024, there were 2 million monthly SDK downloads. OpenAI adopted it in April 2025 — 22 million. Microsoft integrated it into Copilot Studio in July 2025 — 45 million. March 2026 — 97 million. Gartner estimates that by end of 2026, 80% of Fortune 500 companies will have AI agents running in production. Most of those agents connect to tools through MCP.
The PM question here is blunt: Is your product discoverable by AI agents?
Right now, 84% of product leaders say they have "AI integration." But being in someone's Zapier workflow and being visible to an autonomous AI agent are completely different things.
Zapier integration is something the user consciously configures. MCP is something the agent discovers and uses on its own.
When I think about this for clinical SaaS — appointment scheduling, patient records, prescriptions — the gap is obvious. A patient's AI assistant can already book a restaurant, change a flight, draft an email. If the clinic's software has no MCP server, the assistant simply cannot see the clinic's system. It can't check availability. It can't confirm an appointment. It can't send a reminder.
That's not a missing feature. That's invisibility.
So who puts this on the roadmap?
It's easy to label MCP server as "developer infrastructure" and deprioritize it. Exactly what teams did with mobile responsiveness in 2013. "We're desktop-first, mobile isn't a priority." Then customers quietly stopped tolerating products that didn't work on their phones — and the churn was impossible to explain.
The pattern is always the same: user behavior shifts first, tooling catches up, then products that didn't adapt get blindsided by churn they can't attribute.
AI agents are becoming the default working environment for a growing share of users. Being in that environment means being discoverable in it first.
In December 2025, Anthropic donated MCP to the Agentic AI Foundation under the Linux Foundation, co-founded by Anthropic, Block, and OpenAI — with Google, Microsoft, AWS, and Cloudflare as founding supporters. The protocol war is over. MCP won.
At this point, the PM decision isn't "should we build an MCP server?" That ship has sailed.
The real question is: when does this become a roadmap priority?
Do you label it "infrastructure" and slot it behind three other initiatives? Or do you frame it as a distribution channel decision and bring it into strategy conversations?
Whoever handles your API integrations can probably build the MCP server. But first, a PM needs to recognize this as a question about how the product gets discovered — not a question about developer tooling.
97 million installs is enough of a signal to have that conversation.
At least, I think so.