MCP adoption tops 170 organizations; community conference draws about 1,200 attendees

MCP is the plumbing for AI agents — the layer that lets a model talk to tools, files, databases, and apps without every team inventing its own connector. That sounds boring, but it’s the difference between a toy demo and software that actually does work. The news is that MCP now looks less like a clever developer standard and more like real infrastructure. The Agentic AI Foundation’s first North America summit in New York drew about 1,200 people in early April, and the foundation says it has grown past 170 member organizations just a few months after launch. (infoq.com) ### What is MCP, exactly? MCP — Model Context Protocol — started at Anthropic in November 2024 as a standard way for AI systems to connect to external tools and data. Instead of writing a custom integration for every model and every app, developers can expose capabilities through one shared protocol. Basically, it’s a common socket for agent tools. (anthropic.com) ### Why did thi(infoq.com)t looking like industries. The April 2–3 MCP Dev Summit at the New York Marriott Marquis wasn’t just a meetup — it had more than 95 sessions, major cloud and software companies on stage, and a Linux Foundation-backed structure around the protocol. That’s a signal that the ecosystem now expects MCP to be around for a while. (aaif.io)edule/)) ### Why is 170 organizations a big deal? Open standards live or die on adoption. A spec can be elegant and still go nowhere if only one vendor cares. Here, the opposite is happening. AAIF has passed 170 member organizations since its December 2025 launch, and one analyst note called it the fastest-growing foundation in Linux Foundation history. Even if you discount the hype, that is very fast institutional uptake. (sdtimes.com) ### What about the download numbers? The exact number depends on which snapshot you use, but the direction is unmistakable. Anthropic said in February that MCP had reached 100M monthly downloads. More recent ecosystem writeups cite roughly 97M to 110M monthly SDK downloads, while OX Security’s April research framed total exposure at 150M-plus downloads across dependent projects and packages. Those are diff(sdtimes.com)hing: MCP is everywhere. (anthropic.com) ### So why are people suddenly talking about governance? Because once a protocol becomes normal, the risks stop being theoretical. Anthropic donated MCP to the Linux Foundation-backed Agentic AI Foundation in late 2025, which means the project is now being run more like public infrastructure than a vendor-side experiment. That changes the conversation from “can developers use this?” to “who controls changes, naming, compatibility, and trust?” (anthropic.com) ### What broke on the security side? In April, OX Security disclosed what it called an architectural flaw in MCP’s STDIO transport, arguing that the design could enable arbitrary command execution in vulnerable deployments. OX estimated more than 7,000 publicly reachable servers and up to 200,000 exposed instances overall. Anthropic has pushed back on the framing, but the (anthropic.com)come table stakes. (ox.security) ### Why does that change the whole story? Because standards get harder after they win. Early on, MCP’s job was to make integrations easier. Now its job is to make them safe, governable, and boring at enterprise scale. Registries, auth, server identity, transport choices, and policy controls sound like cleanup work — but turns out that’s what maturity looks like. (aai([ox.security)ttom line? MCP has crossed the line from developer convenience to shared infrastructure. The summit attendance and member growth show adoption. The security fight shows the cost of success. If MCP keeps winning, the next phase won’t be about whether agents can connect to tools — it’ll be about who gets trusted to run the connection layer.