The MCP Ecosystem Landscape
An ecosystem at scale
One year after launch, MCP is infrastructure.
In November 2024, Anthropic released MCP as an open standard. By November 2025, over 10,000 active servers existed. Monthly SDK downloads exceeded 97 million. The TypeScript SDK alone sees 5.65 million weekly npm downloads with more than 21,000 dependent projects.
Downloads grew 80x in the first six months—from 100,000 in November 2024 to 8 million by April 2025. Third-party registries now index over 17,000 server implementations.
The MCP Registry
The official MCP Registry launched in preview on September 8, 2025 at registry.modelcontextprotocol.io. It is the central catalog for publicly available MCP servers.
The registry contains nearly 2,000 servers—a 407% increase from the initial batch onboarded at launch. Servers are categorized by function: database access, API integration, developer tools, file systems, and utilities. Each listing includes the server name, description, installation instructions, and source repository.
The registry is itself an MCP resource. Its API uses OpenAPI specification, allowing compatible sub-registries to index and extend its contents. Community moderation keeps quality consistent. Organizations can run private registries for internal servers using the same specification.
The registry maintainer team includes contributors from Anthropic, GitHub, Microsoft, VS Code, and nine other organizations. When that many competitors maintain shared infrastructure together, the infrastructure matters.
Start server discovery at registry.modelcontextprotocol.io. For broader coverage including unofficial servers, directories like mcp.so and Glama.ai index over 17,000 implementations.
Linux Foundation governance
On December 9, 2025, Anthropic donated MCP to the Linux Foundation. The protocol now lives under the Agentic AI Foundation (AAIF), a directed fund for AI agent infrastructure.
Three founding projects populate the AAIF:
Model Context Protocol (from Anthropic) defines the standard for agent-tool communication.
goose (from Block) is an open-source, local-first AI agent framework. Block built it to power their own engineering tools.
AGENTS.md (from OpenAI) establishes a markdown-based standard for AI coding agent guidance. Over 60,000 open-source projects now include AGENTS.md files.
The platinum members: Amazon Web Services, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, and OpenAI. Gold members include Cisco, Datadog, Docker, IBM, JetBrains, Oracle, Salesforce, SAP, Shopify, Snowflake, and Twilio.
Governance works in layers. The AAIF Governing Board handles strategic investments, budget allocation, and member recruitment. Individual projects keep full autonomy over technical direction. MCP continues its existing governance model: maintainers make decisions guided by community input through the Standards Enhancement Process (SEP).
The Linux Foundation provides neutral home and infrastructure. Mike Krieger, Anthropic's Chief Product Officer, explained the donation: "Donating MCP to the Linux Foundation ensures it stays open, neutral, and community-driven as it becomes critical infrastructure."
When competitors jointly fund infrastructure, it signals that the technology has moved from competitive advantage to table stakes. The founding members of AAIF compete intensely on AI models and products while agreeing that the integration layer should be shared.
Major platform adoption
The largest AI platforms adopted MCP throughout 2025.
OpenAI announced MCP support on March 26, 2025. Sam Altman stated: "People love MCP and we are excited to add support across our products." MCP now works with the OpenAI Agents SDK, Responses API, and ChatGPT desktop application. OpenAI became a core contributor to MCP and co-founded the AAIF.
Google confirmed Gemini support on April 9, 2025. Demis Hassabis, CEO of Google DeepMind, declared: "MCP is a good protocol and it's rapidly becoming an open standard for the AI agentic era." At Google I/O 2025, native MCP SDK support arrived for the Gemini API in Python and JavaScript. Google built custom MCP servers for Google Maps and database integrations.
Microsoft previewed MCP support in Windows 11 at Build 2025 on May 19, 2025. GitHub and Microsoft joined the MCP Steering Committee at the same event. By December 2025, Windows 11 Insider Preview Build 26220.7344 shipped with native MCP support.
Windows MCP integration includes:
- The On-device Agent Registry (ODR) for managing local servers
- Built-in connectors for File Explorer and Windows Settings
- Mandatory code signing requirements for MCP server binaries
- Tool poisoning prevention mechanisms
Microsoft's approach is security-first. David Weston, Microsoft VP of Enterprise and OS Security, stated: "As AI agents become more capable and integrated into daily workflows, the need for secure, standardized communication between tools and agents has never been greater."
First-class MCP support now spans Claude, ChatGPT, Gemini, Cursor, VS Code, Microsoft Copilot, and Amazon's Bedrock ecosystem. Any MCP server works with all of them.
Analyst predictions
Gartner's 2026 predictions on vendor adoption:
- 75% of API gateway vendors will have MCP features by 2026
- 50% of iPaaS vendors will have MCP features by 2026
Kong now advertises "industry-leading LLM, MCP, and event streaming support." Google Apigee bundles MCP servers into AI products. Solo.io's Agent Gateway provides MCP and A2A (agent-to-agent) protocol awareness.
Other Gartner predictions for agentic AI:
- 40% of enterprise applications will include task-specific AI agents by end of 2026, up from less than 5% in 2025
- By 2028, 70% of software engineering teams building multimodel applications will use AI gateways
- By 2028, 33% of enterprise software will include agentic RAG (retrieval-augmented generation)
Gartner published multiple reports on MCP: "Innovation Insight: Model Context Protocol," "Emerging Tech: Security Implications of Model Context Protocol," and "Innovation Insight: MCP Gateways." The last identifies MCP gateways as the "missing enterprise layer" for registration, discovery, authentication, and observability.
Analyst predictions drive procurement decisions. When Gartner says 75% of API gateway vendors will have MCP features, enterprise vendors accelerate their roadmaps. This becomes self-fulfilling.
Enterprise deployment examples
Production MCP deployments span industries.
Bloomberg reduced time-to-production from days to minutes after adopting MCP as an organization-wide standard. Sabhav Kothari, Head of AI Productivity, noted: "MCP provides the essential connective layer required in our work building and deploying agentic AI systems for finance." Bloomberg had independently developed an MCP-like protocol internally in early 2024 before switching to the standard.
Block (parent company of Square and Cash App) built goose, an internal AI agent, entirely on MCP architecture. All MCP servers run in-house for complete security control. Block co-founded the AAIF alongside Anthropic and OpenAI.
Amazon uses its API-first culture dating to the mid-2000s. Thousands of internal APIs are now accessible via MCP servers. AWS is a platinum AAIF member.
Microsoft Foundry reported: "We went from a small set of tools to thousands because MCP let tools from GitHub, Azure, and M365 show up wherever agents run."
Ecosystem statistics
Current metrics as of late 2025:
| Metric | Value |
|---|---|
| Monthly SDK downloads | 97+ million |
| Official registry servers | ~2,000 |
| Third-party indexed servers | 17,000+ |
| MCP clients | 300+ |
| npm dependent projects | 21,000+ |
| Discord contributors | 2,900+ |
| New contributors per week | 100+ |
| Active maintainers | 58 |
| Core/lead maintainers | 9 |
| SEPs processed per quarter | 17 |
The most downloaded servers handle common integrations:
| Server | Downloads |
|---|---|
| GitHub MCP | 889,000 |
| Fetch (Anthropic) | 801,000 |
| Context7 | 590,000 |
| Playwright | 590,000 |
| Filesystem (Anthropic) | 575,000 |
Remote MCP server deployments increased nearly 4x since May 2025. Large organizations invest in centralized server infrastructure rather than local-only deployments.
The hype cycle position
MCP moved through the Gartner Hype Cycle faster than typical infrastructure:
Innovation Trigger (November 2024 – February 2025): Anthropic launched MCP. Early adopters experimented.
Peak of Inflated Expectations (March – July 2025): Approximately 25 builders for every actual user. Many servers became "ghost towns"—built but rarely used.
Trough of Disillusionment (Mid-Late 2025): Organizations encountered real-world constraints. Security gaps, missing governance features, and scalability issues surfaced.
Slope of Enlightenment (Late 2025 – Early 2026): Shift toward practical enterprise adoption. Security hardening, governance features, and production deployment patterns matured.
The protocol reached critical mass faster than expected because multiple AI vendors adopted it simultaneously. More clients meant more value for server builders. More servers attracted more clients. Network effects compounded.
Practical implications
Common integrations—GitHub, PostgreSQL, filesystem access, browser automation—already exist. Check the registry before building custom servers.
Claude Code, Cursor, ChatGPT desktop, and VS Code all work with the same servers. Skills transfer across tools.
The Linux Foundation provides long-term stewardship. The protocol is not going away, and changes follow a transparent process.
Bloomberg, Block, and Amazon have production deployments. Security and compliance patterns have been worked out.
The question has shifted from "whether to use MCP" to "which servers to deploy."