Nine months after Anthropic open-sourced the Model Context Protocol in November 2024, the SDK was logging over 97 million monthly downloads, according to Anthropic's own ecosystem update. That is not a niche developer experiment. It is a standard gaining real traction, fast.
The numbers below are drawn from primary sources: Anthropic announcements, the official MCP Registry API, GitHub's search index, and published surveys. Where a figure is self-reported or vendor-cited, that is labeled. Where sources conflict, both figures are shown.
Key takeaways
- The official MCP Registry counted 9,652 server records as of May 2026; Anthropic separately cites more than 10,000 active public servers.
- GitHub returned 15,926 repositories tagged
mcp-serveron the same date, with the canonicalmodelcontextprotocol/serversrepo holding 86,148 stars. - Monthly SDK downloads exceeded 97 million, per Anthropic's cited figures — growth from roughly 100,000 in November 2024.
- 41% of software organizations surveyed by Stacklok in 2026 were in limited or broad production with MCP servers.
- 72% of developers surveyed by Zuplo expect their MCP usage to increase; security is the top-cited blocker.
How fast did MCP server downloads grow?
The SDK download curve is the clearest signal of how quickly MCP moved from announcement to active use. According to figures cited by Anthropic and tracked by Digital Applied, monthly SDK downloads grew from approximately 100,000 in November 2024 to over 8 million by April 2025. By mid-2026, Anthropic cited more than 97 million monthly downloads.
That is roughly an 80x increase in five months, followed by continued compounding. No single figure better illustrates why the MCP ecosystem moved from curiosity to infrastructure concern so quickly.
How many MCP servers exist?
Three independent counts give slightly different answers, which is worth understanding rather than smoothing over.
| Source | Count | Date | Notes |
|---|---|---|---|
| Anthropic ecosystem update | 10,000+ active public servers | December 2025 | Self-reported by Anthropic |
| Official MCP Registry API | 9,652 latest server records | May 24, 2026 | Direct API pull, per Digital Applied |
| Zuplo State of MCP report | 17,000+ publicly listed | 2025/2026 | Zuplo survey report |
GitHub mcp-server topic | 15,926 repositories | May 24, 2026 | GitHub Search API pull |
The divergence reflects methodology. The registry counts formally registered servers. GitHub counts repositories tagged with the topic, including works-in-progress and forks. Zuplo's figure includes aggregators. None of these is wrong — they measure different things. The honest summary: somewhere between 10,000 and 17,000 publicly available MCP servers exist, depending on how you count.
For context, the modelcontextprotocol/servers reference repository had 86,148 stars and 10,799 forks as of the same May 2026 snapshot — one of the faster-growing open-source repositories in the AI infrastructure space.
What does enterprise production adoption actually look like?
Ecosystem size and download counts tell you about developer interest. Production deployment tells you whether organizations are betting real workflows on the protocol.
Stacklok's 2026 software report — cited by Digital Applied as the strongest enterprise survey source available — found that 41% of surveyed software organizations were in limited or broad production with MCP servers. That figure replaces an earlier, widely-circulated 78% production-adoption claim that lacked a traceable primary source.
41% in under two years of a new protocol is a meaningful number. For comparison, most enterprise infrastructure standards take three to five years to reach that level of organizational penetration.
Named enterprise deployments include Block, Bloomberg, and Amazon, per guptadeepak.com's enterprise adoption guide. These are self-reported or press-announced — not independently verified — but they indicate the protocol has moved beyond greenfield experimentation into organizations with compliance and procurement overhead.
Which platforms support MCP?
Platform support is the other half of the adoption story. A protocol only becomes a standard when multiple competing vendors implement it independently.
According to platform documentation and announcements cited by Digital Applied, the following have first-party MCP support or MCP server infrastructure:
- AI model providers: Anthropic (Claude), OpenAI (ChatGPT, Codex), Google
- Development tools: GitHub Copilot, VS Code, Cursor
- Cloud and infrastructure: Microsoft Azure, AWS, Vercel
- Governance: Anthropic donated MCP to the Linux Foundation (now the Agentic AI Foundation, an AAIF project) toward the end of 2025, per Nordic APIs and Zuplo
When OpenAI, Google, and Microsoft all ship first-party support for a protocol Anthropic created, that protocol has crossed the threshold from vendor initiative to open standard. The Linux Foundation donation accelerated that perception.
What are developers actually using MCP for?
The Zuplo State of MCP survey, drawn from nearly 100 technical professionals with hands-on MCP experience, found:
- 72% expect their MCP usage to increase over the coming period
- Security is the number-one blocker to broader production adoption
- Gateways are the top hosting approach for MCP servers in production
The security finding deserves attention. As Nordic APIs notes, MCP adoption has coincided with a documented rise in context window usage and new vulnerability surface area — agents that can call tools can also be manipulated through those tools. This is covered in more depth in MCP Server Security for Enterprise: Keep AI Informed, Not Exposed.
On use cases, the a16z deep dive on MCP published in March 2025 identified the most common patterns as: connecting agents to internal databases, giving coding assistants access to documentation and APIs, and enabling multi-step agentic workflows that span multiple tools. Those patterns map directly to what teams building on MCP report doing in production.
How does the AI agents market frame this growth?
MCP adoption does not happen in isolation. It is accelerating alongside the broader expansion of agentic AI.
According to Precedence Research, cited by Nevermined, the AI agents market was valued at $7.92 billion in 2025 and is projected to reach $236.03 billion by 2034. As agent systems mature, the need for a standard integration layer — which is exactly what MCP provides — grows with them.
The Nevermined analysis of 55 emerging standards adoption statistics frames MCP and Google's A2A protocol as the two leading candidates for how agentic systems will communicate with tools and with each other. The race to establish interoperability standards is the defining infrastructure question of the current agentic moment.
What does MCP adoption mean for teams giving AI real company context?
The numbers above describe a protocol going through standardization. What they imply for operators and engineering teams is more specific.
MCP's value is that it gives any compatible AI agent — Claude, ChatGPT, Cursor, Codex — a consistent way to call tools and retrieve context. But a tool connection is only as useful as the context on the other end. An MCP server pointing at a generic knowledge base gives an agent generic answers. An MCP server pointing at your company's actual data — your Slack threads, your CRM history, your support tickets — gives it answers that are specific to your situation.
That is the gap the adoption curve is now surfacing. As What a Context Layer Gives AI Agents That Bigger Models Cannot explains, model capability is not the bottleneck for most business workflows. Company-specific context is. The MCP adoption statistics show the protocol winning. The next question is what sits behind the server.
For teams evaluating that question, What a Managed Company Brain Looks Like on Your Own Apps and the build vs. buy analysis are practical starting points.
Frequently asked questions
How many MCP servers are there in 2026?
Between 10,000 and 17,000, depending on the counting method. The official MCP Registry API returned 9,652 latest server records in a May 2026 pull. Anthropic cited more than 10,000 active public servers in December 2025. Zuplo's report puts the publicly listed figure above 17,000. GitHub shows 15,926 repositories tagged mcp-server. The range reflects different methodologies, not conflicting data.
What percentage of organizations are using MCP in production?
Stacklok's 2026 software report found 41% of surveyed software organizations in limited or broad production with MCP servers. This is the strongest enterprise survey figure with a traceable primary source. An earlier 78% figure circulated widely but lacks a verifiable origin and should be treated as unsourced.
How fast are MCP SDK downloads growing?
From approximately 100,000 monthly downloads at launch in November 2024 to over 8 million by April 2025, and over 97 million by mid-2026, per figures cited by Anthropic. That represents roughly 80x growth in the first five months, followed by continued compounding.
Which AI tools support MCP?
Anthropic's Claude, OpenAI's ChatGPT and Codex, Google's AI products, GitHub Copilot, VS Code, Cursor, and cloud infrastructure from Microsoft and AWS all have documented first-party MCP support or MCP server infrastructure, per platform announcements cited by Digital Applied.
Is MCP an open standard?
Yes, as of late 2025. Anthropic donated MCP to the Linux Foundation, which houses it under the Agentic AI Foundation (AAIF). This governance move — confirmed by Nordic APIs and Zuplo — is what allowed competing vendors to adopt it without it being perceived as an Anthropic-controlled standard.
What is the biggest barrier to MCP adoption in production?
Security, according to Zuplo's survey of nearly 100 technical professionals. Agents that can call external tools through MCP also expose new attack surfaces — prompt injection through tool responses being the most commonly cited concern. Gateways are the top hosting pattern partly because they allow authentication and rate-limiting to be enforced at the protocol boundary.
How does MCP relate to the broader AI agents market?
MCP is the integration layer that lets AI agents connect to tools and data sources consistently. As the AI agents market grows — Precedence Research projects it from $7.92 billion in 2025 to $236 billion by 2034 — the demand for a standard protocol grows with it. MCP and Google's A2A are currently the leading candidates for how that integration layer gets standardized, per Nevermined's analysis.
Related reading
- MCP Server Security for Enterprise: Keep AI Informed, Not Exposed — the security findings from the Zuplo survey deserve a full treatment; this post covers the attack surfaces and mitigations.
- What a Context Layer Gives AI Agents That Bigger Models Cannot — why the protocol winning does not solve the company-context problem on its own.
- Build vs Buy Context Layer: The Real Cost of Keeping It Running — for teams deciding whether to build their own MCP-backed context layer or use a managed one.
- AI Context Layer Primitives: A Production Evaluation Framework — how to evaluate what sits behind an MCP server before committing to an architecture.
The MCP adoption curve is real and the numbers are now large enough to treat the protocol as infrastructure rather than experiment. If you want your existing AI tools to reach your company's actual data through that infrastructure, start building your company brain with Gyld — connect the apps you already use and expose them as an MCP server your agents can query today.