The Model Context Protocol (MCP) has recorded approximately 97 million monthly SDK downloads as of March 2026, confirming the open standard's position as the primary infrastructure layer for agentic artificial intelligence (AI) applications. This milestone marks a paradigm shift in how AI models connect with external tools, databases, and APIs.
From Zero to 97 Million: The Meteoric Rise of MCP
Anthropic open-sourced MCP on November 25, 2024, alongside reference servers for Google Drive, Slack, Github, and Postgres, with native support built into Claude Desktop. Early adopters included Block and Apollo; IDEs such as Zed, Replit, Codeium, and Sourcegraph began integration within weeks. The protocol defines how AI models connect to external tools, databases, APIs, and workflows. An MCP host, such as Claude Desktop, ChatGPT, or VS Code Copilot, communicates with an MCP server—a lightweight wrapper around a specific tool or data source. One server can serve every compatible client without per-model custom code.
The official project site describes it as "a USB-C port for AI applications." That framing captures the practical shift: instead of building separate connectors for each AI platform, developers expose a single MCP server and gain compatibility across Claude, ChatGPT, Gemini, Microsoft Copilot, and any other MCP-compatible client.
Crypto Sector Embraces MCP Infrastructure
OpenAI added full MCP support across ChatGPT and its Agents SDK in March 2025, which analysts identified as an inflection point in adoption. Google, Microsoft, AWS, and dozens of other platforms followed through mid-2025. In December 2025, Anthropic donated MCP to the newly formed Agentic AI Foundation (AAIF) under the Linux Foundation. OpenAI and Block joined as co-founders. Platinum members include AWS, Google, Microsoft, Cloudflare, Github, and Bloomberg. The governance structure mirrors that of Kubernetes and Pytorch—vendor-neutral and community-managed.
As of March 2026, more than 10,000 active MCP servers exist across public and enterprise deployments. The combined Python and Typescript SDKs account for the 97 million monthly download figure, up from roughly 100,000 at launch in late 2024. The crypto sector has moved quickly to build MCP infrastructure. BitGo launched an official MCP server in March 2026, enabling AI tools and development environments to interact with its institutional digital asset custody platform via natural language.
Coinbase released a Payments MCP through its Developer Platform in late 2025, connecting AI agents to crypto wallets, onramps, and stablecoin transactions. Crypto.com released a Market Data MCP server providing live price quotes, order books, and candlestick charts. CoinGecko launched its own server supplying real-time data for more than 15,000 cryptocurrencies and 1,000-plus exchanges. Cross-chain protocol deBridge deployed an MCP server in February 2026, enabling non-custodial swaps and bridging across EVM chains and Solana.
Enterprises building internal AI tooling have shifted away from one-off API connectors. Publishing a single MCP server gives a product instant interoperability with every major AI client—a network effect that proprietary integration approaches cannot replicate.
Security Challenges and Governance Progress
Security researchers have flagged that many public MCP servers have not undergone formal audits. The Linux Foundation stewardship brings standardized authentication and transport requirements, but deployment-level security remains the responsibility of individual server maintainers. The protocol reached that comparison point in under 18 months. The MCP specification, SDKs, and reference implementations are available at github.com/modelcontextprotocol. The core documentation lives at modelcontextprotocol.io.
"MCP reached a comparison point in under 18 months. It demonstrates the immense potential for AI and blockchain infrastructure convergence." — Industry analyst comment.