Model Context Protocol Hits 97 Million Monthly Downloads — AI's New Infrastructure Layer

The Model Context Protocol crossed 97 million monthly SDK downloads in March 2026. Anthropic launched it in November 2024 with approximately 2 million downloads. That’s a 4,750% growth rate in 16 months.

Those numbers mean MCP has crossed the threshold that separates interesting experiments from actual infrastructure. When an open standard hits 97 million downloads and every major AI provider has committed to it, it stops being optional.

Why it spread. MCP standardizes the interface between AI agents and external tools — how an agent discovers what tools exist, how it calls them, how it handles errors. Before MCP, every AI framework had its own tool format. OpenAI’s function calling, Anthropic’s tool use, Google’s function declarations — each required separate integration work for every tool. MCP replaced that fragmentation with a single protocol. Build an MCP server once, and it works with every compliant agent.

The ecosystem. Over 5,800 community and enterprise MCP servers now exist covering databases, CRMs, cloud providers (AWS, GCP, Azure), productivity tools, developer tools, e-commerce platforms, and analytics services. Block, Bloomberg, Amazon, and hundreds of Fortune 500 companies have production MCP deployments. Anthropic donated the protocol to the Linux Foundation’s Agentic AI Foundation in December 2025, with AWS, Bloomberg, Cloudflare, Google, and Microsoft as platinum members.

What’s still missing. Despite the adoption, three critical gaps remain. MCP has no built-in cost attribution or token counting — CTO of Merge Gil Feig estimates tool metadata overhead accounts for 40–50% of available context in typical deployments. There’s no standardized identity propagation, which makes audit trails and SSO-integrated authentication difficult in enterprise settings. And structured error semantics are absent, meaning agents can’t reliably distinguish between “tool unavailable” and “operation failed.”

The 2026 roadmap targets these gaps, with enterprise authentication and a formal registry expected to land this year.

The comparison that keeps coming up is REST APIs. REST didn’t win because it was technically superior to SOAP — it won because it was simple enough that everyone adopted it, and once everyone adopted it, you couldn’t ship a web product without supporting it. MCP is doing the same thing for AI agents. The 97 million downloads aren’t the story. The story is that 5,800 servers mean there’s almost certainly already an MCP server for whatever you’re trying to connect — you don’t have to build the integration anymore, you just have to ship the agent.

Sources: MCP Roadmap 2026, Digital Applied