MCP hits 97M downloads — Anthropic just gave it away

In December 2025, Anthropic made a move that could define how all AI tools work together for the next decade: it donated the Model Context Protocol (MCP) — the "USB standard for AI tools" — to a neutral foundation backed by every major technology company on Earth. As of March 2026, the protocol has reached 97 million monthly downloads, with adoption growing faster than any AI standard in history.

If you use Claude, ChatGPT, Cursor, or any AI assistant that can access outside tools, there's a good chance MCP is the invisible glue making it work. And now, no single company owns it.

What MCP Actually Does (No Jargon Version)

Before MCP existed, connecting an AI assistant to an external tool — say, your company's calendar app or document system — required custom engineering work every single time. An AI company and a software company had to build a bespoke "integration" (a custom-built bridge between two software systems) between their products.

With 10 AI products and 10 software tools, that means 100 separate integrations. With 100 of each? 10,000. Engineers call this the "n×m integration problem" — and it was becoming a serious bottleneck for everyone building with AI.

Instead of building 100 separate integrations, developers now build one MCP server (a small program that exposes a tool's capabilities in a standard format) and it instantly works with every compatible AI platform.

The Scale: 14 Months, 97 Million Downloads

MCP was introduced just 14 months ago — a timeline so short it's almost hard to believe given the adoption numbers today:

• 97 million monthly SDK downloads across Python and TypeScript packages
• 10,000+ public MCP servers globally deployed and accessible
• 82,300 GitHub stars on the main servers repository — accumulated in under 8 months
• 45,000+ followers on the MCP GitHub organization
• 75+ pre-built connectors in Claude's official connector directory
• 10+ official programming language SDKs: Python, TypeScript, Java, Kotlin, C#, Go, PHP, Ruby, Rust, and Swift (the C# SDK was built in collaboration with Microsoft)

The broader AI developer ecosystem puts this in context: 1.13 million public code repositories now import AI model libraries — up 178% year-over-year. In 2026 alone, 693,000 new AI repositories were created. MCP sits at the center of this explosion as the standard protocol that lets all these tools talk to each other.

Who's Behind It Now? Seven Major Organizations.

Anthropic donated MCP to the Agentic AI Foundation (AAIF), a new directed fund established inside the Linux Foundation — the non-profit organization that governs Linux, Kubernetes (the software that powers most of the world's cloud computing), and thousands of other critical technology standards.

The founding and supporting members represent a rare moment of industry-wide alignment:

All major AI platforms already support MCP natively: ChatGPT, Claude, Cursor, Gemini, Microsoft Copilot, and VS Code. This is not experimental adoption — it is the active standard that the world's leading AI tools are built on right now.

What Neutral Governance Actually Means for You

Before this move, MCP was controlled entirely by Anthropic. One company could change the rules, add restrictions, or deprecate (retire and remove) parts of the protocol at any time. Moving to the Linux Foundation changes the power structure entirely:

• No single company can change the rules — governance is now shared across all member organizations, similar to how no company "owns" HTTP (the protocol that makes websites work)
• Enterprise-grade security standards will be formalized into the spec — businesses can build on MCP knowing it has the stability of a Linux Foundation project
• Compatibility guarantees — tools built on MCP today will still work in 5 years, because no single company can quietly break backwards compatibility
• Equal participation rights — a small startup can contribute to the MCP standard on equal footing with Google or Microsoft

The closest historical analogy: this is the moment MCP stopped being "Anthropic's protocol" and became shared infrastructure — the same transition that happened when TCP/IP (the foundational protocol that routes all data across the internet) moved to an open standards body and became the permanent foundation of the web.

Get Started — One Command

Official SDKs are available in 10+ languages. Install and start building in under a minute:

# Python SDK
pip install mcp

# TypeScript/Node.js SDK
npm install @modelcontextprotocol/sdk

# Test your MCP server visually in a browser
npx @modelcontextprotocol/inspector

The official MCP specification covers five core capabilities: Resources (files and data the AI can read), Prompts (reusable instruction templates), Tools (actions the AI can take), Sampling (requesting help from another AI), and Elicitation (the AI requesting more input from the user). Recent additions include Tool Search, async (background, non-blocking) operations, and server identity verification for enterprise security.

Not a developer? If you use any AI assistant that connects to your calendar, documents, or work apps — MCP is almost certainly already running underneath. What this news means for you: the protocol powering your AI tools is now governed by a neutral foundation with a long-term commitment to stability. The tools you build workflows around today will keep working.

Browse the full catalog of 10,000+ MCP connectors at github.com/modelcontextprotocol/servers, or read the official docs at modelcontextprotocol.io.

Related Content — Get Started with Easy Claude Code | Free Learning Guides | More AI News