From Experimental Spec to Infrastructure Layer
Anthropic announced that its Model Context Protocol (MCP) crossed 97 million monthly SDK downloads on March 25, 2026. For a spec that was an internal Anthropic experiment at launch, that number represents one of the fastest adoption curves for any developer infrastructure standard on record. Independent coverage frames the number as the moment MCP formally transitioned from experimental Anthropic-specific tech to the foundation layer of agentic AI across the industry.
MCP is, at its core, a universal interface between AI models and external tools, data sources, and APIs. Instead of each provider (Claude, GPT, Gemini) implementing bespoke tool integrations, a developer builds an MCP server once, and any MCP-compatible agent can call it. The simplicity of that contract is what let the protocol spread.
The Numbers Behind the Adoption Curve
According to Anthropic and corroborated by Effloow, Vucense, and byteiota, the ecosystem by late March 2026 included:
- 97 million monthly combined Python + TypeScript SDK downloads (up from roughly 2 million at launch)
- 10,000+ public MCP servers indexed across registries
- 300+ MCP clients across code editors, chat apps, and enterprise platforms
- 4x growth in remote MCP server deployments since May 2025
- 72% of developers already using MCP plan to increase usage
That is not vendor marketing; those numbers are from a protocol that is now effectively infrastructure. Every major AI provider — OpenAI, Google DeepMind, Microsoft, Meta — ships MCP-compatible tooling as default, removing the reason to build a competing spec.
Advertisement
Governance Moved to the Linux Foundation
On December 9, 2025, Anthropic donated MCP to the newly formed Agentic AI Foundation (AAIF) under the Linux Foundation. The foundation was co-founded by Anthropic, Block, and OpenAI. AWS, Google, Microsoft, Cloudflare, and Bloomberg joined as platinum members.
That move mattered more than the adoption chart. Protocols tied to a single vendor — no matter how popular — struggle to become durable standards because competitors have a permanent incentive to fork. Anthropic’s decision to hand MCP to a neutral foundation removed that incentive, and its biggest competitors became its biggest foundation partners. For enterprises evaluating whether to standardize on MCP, that governance move effectively de-risked the bet.
What MCP Winning Unlocks for Builders
The practical implication: when engineers are choosing how to give an AI agent access to a database, a CRM, a ticketing system, or a private API, the answer in 2026 is “build or consume an MCP server.” The 2026 roadmap, published on blog.modelcontextprotocol.io and analyzed by The New Stack, prioritizes four areas: transport scalability, agent-to-agent communication, governance maturation, and enterprise readiness.
Forrester forecasts that 30% of enterprise app vendors will ship their own MCP servers, turning the protocol into a customer-acquisition surface: if an enterprise’s agent already speaks MCP, a SaaS vendor that exposes an MCP server is one wiring step away from being integrated. For API-first companies, that shifts the unit of distribution from SDKs and webhooks to MCP.
What to Watch Next
Three fronts will define how the 97M figure evolves. First, authentication and authorization: CData, WorkOS, and the 2026 roadmap all flag identity and fine-grained permissions as MCP’s biggest enterprise gap. Second, agent-to-agent interop: the roadmap commits to A2A communication, which moves MCP from “tool calling” to “multi-agent orchestration.” Third, security: widespread MCP servers are a new attack surface, and the community has already flagged prompt-injection and tool-poisoning risks that enterprise adopters need to harden against.
For builders, the protocol war being over is not the end — it is the start. The question is no longer “which standard?” but “what do I build on top of it?”
Frequently Asked Questions
What exactly are the 97 million MCP installs?
They are monthly combined Python and TypeScript SDK downloads as measured by Anthropic and reported on March 25, 2026 — not a cumulative install count. That number grew from roughly 2 million in the early months, making MCP one of the fastest-adopted developer infrastructure standards on record.
Who controls MCP now — is it still an Anthropic spec?
No. On December 9, 2025, Anthropic donated MCP to the Agentic AI Foundation (AAIF) under the Linux Foundation. The foundation was co-founded by Anthropic, Block, and OpenAI, with AWS, Google, Microsoft, Cloudflare, and Bloomberg as platinum members. MCP is now governed as a neutral industry standard, which is why competing AI vendors have been willing to ship MCP-compatible tooling.
Should a team building AI agents today commit to MCP?
For most new agent work in 2026, yes. Every major AI provider ships MCP-compatible tooling, there are 10,000+ public MCP servers, and the Linux Foundation governance makes long-term fragmentation unlikely. The main caveats are enterprise-grade authentication and security — these are priorities on the 2026 roadmap but still evolving, so teams handling sensitive data should harden access controls around their MCP servers.
Sources & Further Reading
- Donating the Model Context Protocol and establishing the Agentic AI Foundation — Anthropic
- Anthropic’s Model Context Protocol Hits 97 Million Installs on March 25 — AI Unfiltered
- MCP Ecosystem in 2026: From Experiment to 97 Million Installs — Effloow
- A Year of MCP: From Internal Experiment to Industry Standard — Pento
- The 2026 MCP Roadmap — Model Context Protocol Blog
- MCP’s biggest growing pains for production use will soon be solved — The New Stack
- Model Context Protocol — Wikipedia
















