The Model Context Protocol (MCP) is rapidly emerging as a foundational layer for agentic AI, and its latest growth figures suggest the standard has moved well beyond the experimental stage. As of March 2026, the combined monthly downloads of the Python and TypeScript MCP SDKs reached approximately 97 million, underscoring how widely the protocol is being adopted across AI tooling, developer environments, and enterprise systems.
MCP was open-sourced by Anthropic on November 25, 2024, together with reference servers for Google Drive, Slack, Github, and Postgres, plus native support inside Claude Desktop. From the start, the protocol was designed to solve a growing problem in AI software: how to connect language models and autonomous agents to external tools, databases, APIs, and workflows without forcing developers to build separate integrations for each model or platform.
A Standard Interface for AI Tools
MCP is often described as a kind of “USB-C for AI applications”. The comparison is useful because it captures the protocol’s core value proposition: one interface, many compatible endpoints. In practice, an MCP host such as Claude Desktop, ChatGPT, or VS Code Copilot can communicate with an MCP server, which acts as a lightweight wrapper around a specific tool or data source. Instead of writing custom connectors for every individual model provider, a developer can expose a single MCP server and gain compatibility across a growing list of AI clients.
That interoperability has become one of the strongest drivers of adoption. Early users included Block and Apollo, while development environments such as Zed, Replit, Codeium, and Sourcegraph started integrating MCP within weeks of its release. The protocol’s open nature made it especially attractive to teams that wanted broad compatibility without being locked into a single AI vendor’s stack.
Why 2025 Became the Inflection Point
Analysts have pointed to March 2025 as a major turning point for MCP adoption, when OpenAI added full MCP support across ChatGPT and its Agents SDK. That move helped legitimize the protocol beyond Anthropic’s own ecosystem and signaled that major AI platforms were willing to converge on a shared standard. Through the middle of 2025, companies including Google, Microsoft, and AWS followed with support of their own, accelerating MCP’s path toward becoming a de facto interoperability layer for AI agents.
The governance story also evolved quickly. 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, while platinum members included AWS, Google, Microsoft, Cloudflare, Github, and Bloomberg. That structure matters because it shifts MCP away from any single-company narrative and toward a vendor-neutral model, closer to the governance approach used by projects like Kubernetes and PyTorch.
By March 2026, the number of active MCP servers across public and enterprise deployments had surpassed 10,000. Compared with roughly 100,000 SDK downloads around the protocol’s launch in late 2024, the growth curve has been dramatic.
Crypto’s Fast Adoption of MCP
The cryptocurrency industry has moved particularly quickly to build on MCP, treating it not just as an AI protocol but as a practical infrastructure layer for connecting digital asset services to intelligent agents. That trend is important because crypto products often rely on fragmented APIs, multiple data providers, and complex transaction workflows—exactly the kind of environment where standardization can reduce friction.
Among the most notable moves, BitGo launched an official MCP server in March 2026. The release enables AI tools and development environments to interact with BitGo’s institutional digital asset custody platform through natural language interfaces. For institutional users, that could streamline the way operational teams query systems or connect internal AI workflows to custody infrastructure.
Coinbase had already entered the space in late 2025 by releasing a Payments MCP through its Developer Platform. That server connects AI agents to crypto wallets, fiat onramps, and stablecoin transactions, showing how MCP can support financial and payments use cases in addition to pure data access. In effect, Coinbase positioned MCP as a bridge between AI automation and programmable crypto payments.
Crypto.com also introduced a Market Data MCP server, giving developers and AI systems access to live price quotes, order books, and candlestick chart data. Meanwhile, CoinGecko launched its own MCP server delivering real-time market information covering more than 15,000 cryptocurrencies and over 1,000 exchanges. These launches show how market data providers are using MCP to make their information more accessible to AI-native applications.
Cross-chain protocol deBridge joined the trend in February 2026, deploying an MCP server that supports non-custodial swaps and bridging across EVM chains and Solana. That is especially notable because it extends MCP usage from information retrieval into actual blockchain actions, highlighting the protocol’s relevance for agent-driven execution as well as analysis.
Why Enterprises Are Paying Attention
For enterprises building internal AI systems, the appeal of MCP goes beyond convenience. Organizations have increasingly moved away from one-off API connectors because those approaches are expensive to maintain and difficult to scale across multiple AI clients. Publishing a single MCP server offers immediate interoperability with major AI platforms, creating a network effect that proprietary integrations struggle to replicate.
This dynamic may be especially powerful in sectors like crypto, where products span custody, payments, analytics, trading infrastructure, and cross-chain activity. A common protocol can simplify how AI assistants and automated agents interact with these services, potentially making both internal workflows and user-facing experiences more efficient.
Security Remains an Open Challenge
Despite the momentum, security remains a significant concern. Researchers have warned that many public MCP servers have not undergone formal audits. While Linux Foundation stewardship is expected to improve standardization around authentication and transport requirements, deployment-level security still depends on the individual maintainers operating each server.
That distinction is critical. A protocol can provide a common framework, but the trustworthiness of each implementation still varies. As MCP adoption expands into higher-stakes environments—including enterprise systems and crypto infrastructure—questions around server hardening, permissions, authentication, and operational oversight are likely to become more important.
An Infrastructure Layer to Watch
MCP’s rise has been unusually fast. In less than 18 months, it has gone from a newly open-sourced protocol to a broadly recognized standard for AI tool connectivity. The combination of strong developer adoption, support from major AI platforms, and increasing usage by crypto firms suggests MCP is becoming more than just a technical specification—it is turning into a core interoperability layer for the next generation of software agents.
For the crypto industry, that shift could be especially meaningful. As more exchanges, custodians, data providers, and cross-chain protocols expose MCP servers, AI systems may gain a more standardized way to access market information, initiate transactions, and interact with digital asset infrastructure. The next stage of growth will likely depend not only on adoption, but on whether the ecosystem can match its rapid expansion with robust security and reliable governance.