MCP Hits 97 Million Installs as It Becomes Standard AI Agent Infrastructure
Anthropic's Model Context Protocol crossed 97 million monthly SDK downloads on March 25, making it the fastest-adopted AI infrastructure standard in history. OpenAI, Google, Microsoft, and Mistral all integrated MCP support by mid-March 2026.
Anthropic’s Model Context Protocol reached 97 million monthly SDK downloads on March 25, 2026. Kubernetes took nearly four years to hit comparable enterprise deployment density. MCP did it in roughly 15 months.
The number matters because of what it represents: MCP has crossed from “interesting Anthropic project” to assumed infrastructure. The protocol standardizes how AI agents communicate with external tools and data sources — think of it as a universal connector layer that lets any AI model call any tool without custom integration code. If an agent can speak MCP, it can work with any MCP-compatible server. As of mid-March, the client list includes ChatGPT, Claude, Cursor, Gemini, Microsoft Copilot, and Visual Studio Code.
The adoption inflection point coincided with Fortune 500 companies moving agentic AI out of pilot programs and into production. When enterprises start shipping real agentic workflows, they standardize on protocols that actually have ecosystem support. MCP was there at the right time, with the right design.
The governance structure formalized in December 2025 when Anthropic donated MCP to the Agentic AI Foundation (AAIF), a directed fund under the Linux Foundation. The foundation was co-founded by Anthropic, Block, and OpenAI — three companies that were competing on models while agreeing on infrastructure. Google, Microsoft, AWS, Cloudflare, and Bloomberg joined as supporting members. The AGENTS.md spec from OpenAI and Block’s goose framework are also housed under AAIF.
For developers, this is a meaningful signal. A protocol controlled by a neutral foundation with cross-vendor backing is one you can build on without betting on a single company’s product roadmap. The 10,000+ active MCP servers already deployed cover databases, file systems, code execution environments, web browsers, calendar systems, and more. The ecosystem is real enough that picking any other protocol for new agentic work requires a strong reason.
The 97 million figure is installs, not active deployments. Those aren’t the same. But adoption at this scale means documentation is thorough, edge cases are documented on Stack Overflow, and you won’t be the first person to hit whatever problem you encounter. That’s the threshold that matters for production use.
If you’re building anything involving AI agents in 2026 and haven’t implemented MCP yet, the question is no longer whether to — it’s when.