Model Context Protocol Crosses 97 Million Installs — The Infrastructure Milestone Nobody Is Talking About

The Protocol That Just Won

In late March 2026, Anthropic's public dashboards for the Model Context Protocol ecosystem crossed 97 million cumulative installs. The milestone received almost no mainstream press coverage. It received moderate coverage in the developer-focused technical press. It received essentially no executive-level analysis — the CFO playbooks, enterprise strategy think pieces, and investor letters that dominated the coverage of the autonomous coworker transition went right past the infrastructure milestone that was enabling it.

This is a characteristic of how infrastructure protocols get reported. The consequence they enable gets the coverage. The protocol itself gets ignored.

Why The Milestone Matters

Ninety-seven million installs for an enterprise protocol in eighteen months is not a linear-growth outcome. It is a critical-mass outcome, and it means specific things for the competitive dynamics of enterprise AI through 2026 and 2027.

Model providers are no longer differentiated on integration footprint. Anthropic, OpenAI, Google, and Microsoft all now support MCP natively in their enterprise products. Before this year, the most compelling reason to commit to Claude Cowork instead of ChatGPT Enterprise was Anthropic's deeper integration maturity. That differentiation has evaporated. The competition has shifted decisively from "who has the best integrations" to "who has the best model, best reliability, and best go-to-market for our industry."

Enterprise software vendors are standardized on a single interoperation point. Salesforce, ServiceNow, Workday, SAP, Snowflake, Databricks, Microsoft 365, Google Workspace, and roughly 260 other enterprise software vendors have shipped MCP servers. Enterprise customers can now deploy agent platforms against their existing tool stacks without rebuilding integrations. The TCO of agent deployment has dropped dramatically in the last two quarters as a result.

The competitive threat to MCP has effectively collapsed. Google's Agent-to-Agent protocol, OpenAI's function calling, Microsoft's Semantic Kernel, and several smaller initiatives were all credible alternatives in 2024 and early 2025. By April 2026, none of them has the vendor-adoption breadth that MCP has. Protocol competitions of this kind usually end with a single winner. MCP has won.

The Adoption Curve In Detail

The curve tells the story. Slow early growth through Q1 2025 as early adopters validated the protocol. Accelerating adoption through Q2 and Q3 as enterprise pilots proved value. Exponential growth starting in Q4 2025 as standardization pressure from customers forced vendors to ship MCP servers as a competitive necessity. The Q1 2026 jump to 97 million is the signature of a protocol moving from optional to mandatory.

What This Means For Enterprise Buyers This Quarter

Three concrete implications for technology leaders evaluating AI vendors in Q2 2026.

First, MCP compatibility is now a baseline contract requirement. Any enterprise AI vendor that does not ship MCP support for your existing tool stack should be treated as high- risk. The deployment friction of working around an MCP- incompatible vendor is substantial, and the strategic risk of committing to a protocol-specific vendor that later has to retrofit MCP support is real.

Second, the agent-platform vendor decision has a shorter half-life than it did six months ago. Because MCP has abstracted away the integration layer, switching between agent platforms is substantially less expensive than it was in 2025. Enterprises that over-committed to a single vendor's ecosystem should be reassessing whether the architectural lock-in still exists. In most cases it does not.

Third, the security and governance questions about MCP are becoming load-bearing. As more of the enterprise-AI infrastructure runs through MCP, the consequences of MCP-level security vulnerabilities scale accordingly. Enterprise security teams should be paying specific attention to MCP server permissioning, audit logging, and vendor security review processes. The protocol's maturity in these areas is improving but not yet at the level that critical enterprise infrastructure demands.

The Historical Pattern This Resembles

Enterprise protocols that reach critical-mass standardization tend to follow a predictable arc over the subsequent five to ten years. The protocol becomes the default, commercial competition shifts to layers above it, alternative protocols fade or become niche, and the governance of the winning protocol eventually transitions from vendor-led to institution-led.

HTTP followed this arc from 1991 to roughly 2000. OAuth followed it from 2007 to about 2015. REST followed it from 2000 to roughly 2012. Each case took five to ten years. MCP has accelerated the critical-mass phase substantially — reaching in eighteen months what HTTP took four years to achieve — which suggests the subsequent phases may also compress.

Expect the first serious push for MCP governance transition to an independent standards body (IETF or W3C) within the next twelve months. Expect the first major security incidents tied to MCP server vulnerabilities within the next six months. Expect the competitive advantage that Anthropic gains from MCP authorship to gradually diffuse as the protocol's governance democratizes.

For broader context on why MCP matters structurally, see the full analysis of MCP as enterprise agent infrastructure. For the capability milestone MCP enabled, see The Autonomous Coworker Threshold. For how the standardization is affecting enterprise workforce decisions, see Snap's 16% layoffs as canary signal.