Anthropic shipped the Model Context Protocol in November 2024 as an open-source standard for connecting AI models to external tools and data sources. Within a year, every major competitor adopted it. The protocol is now the de facto standard for the agentic AI category — the layer that lets an AI assistant read a file, run a tool, query a database, or talk to another application on the user's behalf.
By EPR Editorial Team · July 2026
The adoption pattern is unusual in AI infrastructure. Standards published by frontier labs typically get ignored by competitors as a matter of strategic positioning. Anthropic's MCP did not. The reasons are observable in the design — the protocol is genuinely open, the reference implementation works, and the early integrations created enough developer pull to make competing standards economically unviable. This is the reference map of every major company that has adopted MCP, what they have integrated, and what it tells you about where the agentic AI stack is heading.
What MCP Is
The Model Context Protocol is an open standard, with reference implementations in TypeScript and Python, that defines how an AI assistant connects to external tools, data sources, and applications. The protocol formalizes three primitives — resources (data the assistant can read), tools (functions the assistant can call), and prompts (templated workflows) — and a transport layer that handles the connection between an MCP client (the AI assistant) and one or more MCP servers (the external integrations).
Before MCP, every AI assistant ran its own proprietary tool-use protocol. OpenAI had function calling. Anthropic had tool use. Google had its own function-calling format. Each was incompatible with the others, and every integration between an AI assistant and an external application had to be built three times. MCP fixed that by publishing one standard that all the assistants could speak. The economic logic is the same logic that drove HTTP adoption against competing application protocols in the 1990s. Standardization at the integration layer benefits everyone in the ecosystem, even the company that did not publish the standard.
The Foundation-Model Adopters
OpenAI announced MCP support in March 2025, beginning with the Agents SDK and extending across the broader OpenAI tooling. The adoption was the single largest signal that MCP had moved from Anthropic-published standard to industry-default. OpenAI's position — the largest AI assistant by consumer scale, the most aggressive product velocity, and the company with the strongest commercial incentive to publish its own competing standard — made the MCP adoption read as a structural acknowledgment that the protocol had won the early developer mindshare.
Google DeepMind followed within weeks of OpenAI. Gemini integrated MCP support across the Workspace, Android, and developer surfaces. The Google adoption locked in the second of the three major foundation-model providers, leaving the integration layer effectively unified across Claude, ChatGPT, and Gemini for the first time.
Microsoft integrated MCP into Copilot Studio, the enterprise platform that builds custom AI assistants on top of the underlying foundation models. The Microsoft adoption brought MCP into the largest enterprise software install base in technology and made the protocol the integration standard for the broad Microsoft enterprise customer set.
The AI coding category was the first to adopt MCP at scale, because the use case — connecting an AI coding assistant to the local file system, the running terminal, the database, the version control system, and the production logs — maps cleanly to the MCP architecture.
Cursor, the AI-first code editor that has reshaped the developer-tools category, integrated MCP servers in early 2025. Claude Code, Anthropic's own terminal-native coding agent, ships with MCP as the primary integration mechanism. Continue, the open-source coding assistant for VS Code and JetBrains IDEs, was an early MCP adopter and contributor. Cline, Aider, Zed, and Codeium have all integrated MCP-compatible server support. The AI coding stack has effectively standardized on MCP for tool integration.
Replit integrated MCP into its browser-based development environment, allowing the platform's AI assistant to connect to the broader ecosystem of MCP servers. Vercel announced MCP support across the v0 generative UI tool. The build-with-AI category has consolidated around the protocol.
The Enterprise SaaS Adopters
The MCP server directory now lists official integrations from a long roster of enterprise SaaS companies. The list reads as a directory of the modern B2B software stack.
GitHub published an MCP server in early 2025, allowing AI assistants to read repositories, query pull requests, and act on issues. Linear shipped an MCP server for project management integration. Notion's MCP server lets AI assistants read and write into Notion workspaces. Slack and Atlassian (Confluence, Jira) shipped MCP servers for communications and ticket-system integration.
The financial-infrastructure layer adopted quickly. Stripe shipped an MCP server for payments operations and customer record queries. Block (formerly Square) shipped integrations for the Cash App and Square commerce surfaces. Sentry built an MCP server for error monitoring and observability. The customer-data-platform layer has been an active MCP server adopter.
Asana, Pinterest, Zed, Cloudflare, Sourcegraph, Apollo GraphQL, Postgres, MongoDB, Snowflake, Databricks, and a long list of additional vendors either ship official MCP servers or have endorsed the protocol. The depth of the enterprise SaaS adoption is the part of the MCP story that is least appreciated in mainstream AI coverage. The protocol has effectively unified the integration layer across most of the working B2B software stack.
The Cloud-Provider Adopters
Cloudflare published a substantial MCP integration that allows AI assistants to deploy, manage, and observe Cloudflare Workers — the company's edge-compute layer — through the standard protocol. The integration is one of the most-cited reference implementations of MCP server design and has been pointed to by Anthropic and by independent developers as the template for production-grade MCP server architecture.
Microsoft, beyond the Copilot Studio integration, exposes MCP across the broader Azure AI Foundry surface. AWS has integrated MCP server compatibility into the Bedrock agent surface, allowing assistants running on Claude (the primary Anthropic distribution channel through AWS) to connect to MCP servers natively. Google Cloud integrated MCP into the Vertex AI Agent Builder.
Why This Matters
Three reasons the MCP adoption pattern is the most important AI infrastructure story of the last 24 months, even though it has received a fraction of the coverage that consumer AI products have.
First, MCP is the integration layer for the agentic AI category. Every agentic AI use case — an AI assistant booking a meeting, an AI assistant filing an expense report, an AI assistant running a customer service workflow — depends on a protocol that connects the assistant to the external systems. MCP is now that protocol. The companies that ship MCP servers are positioned in the agentic AI distribution layer. The companies that have not are positioned outside it.
Second, MCP standardization is the structural reason agentic AI is moving faster than any previous AI capability category. The economics of integration — the cost of building once and running everywhere — compounds. The agentic AI tooling shipping in 2025 and 2026 is the most rapidly scaling category in software, and MCP is the structural reason.
Third, Anthropic published MCP. The protocol is one of the few cases in modern technology where the company that originated the standard also retained commercial leadership in the underlying market. The Anthropic MCP play is studyable as a strategic-PR case — the company published an open standard, allowed competitors to adopt it, and gained more from the ecosystem-wide adoption than it would have gained from a proprietary moat. Communications teams in technology categories where standards-setting is on the strategic table should study how Anthropic did it.
The Adoption Timeline
November 2024: Anthropic publishes the Model Context Protocol with TypeScript and Python reference implementations and the first MCP servers (filesystem, GitHub, Postgres, Slack).
December 2024 through January 2025: Independent developer adoption surges. The MCP server directory grows from the initial Anthropic-published servers to dozens of community contributions across categories.
Early 2025: Cursor, Continue, Cline, Aider, and the AI coding category adopt MCP as the primary tool-integration mechanism. Cloudflare publishes the reference production-grade MCP server implementation.
March 2025: OpenAI announces MCP support across the Agents SDK and ChatGPT tooling. The adoption is the most consequential single signal of MCP becoming the industry standard.
April 2025: Google DeepMind announces MCP support across Gemini. Microsoft integrates MCP into Copilot Studio.
Through 2025: The enterprise SaaS layer — GitHub, Linear, Notion, Slack, Atlassian, Stripe, Block, Asana, Sentry, Pinterest, Zed — ships official MCP servers.
2026: The protocol is the de facto standard. New AI tool integrations ship MCP-first. Competing proprietary protocols continue to exist as legacy infrastructure but new integrations consolidate on MCP.
