Model Context Protocol (MCP): Which Companies Adopted It

The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to external tools and data through one interface. Anthropic introduced MCP on November 25, 2024, and donated it to the Linux Foundation's Agentic AI Foundation on December 9, 2025. Anthropic reports more than 10,000 active public MCP servers and support in ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code.

Updated October 10, 2026. Sources: Anthropic's December 2025 MCP announcement and the official MCP documentation.

What is the Model Context Protocol?

The Model Context Protocol is an open-source standard for connecting AI applications to external systems such as files, databases, search engines and workflows. An MCP client (the AI assistant) talks to one or more MCP servers, and each server exposes a single capability, such as Slack messages or GitHub issues.

Before MCP, every connection between a model and a tool was a custom integration. Anthropic shipped tool use, OpenAI shipped function calling, and Google shipped its own function-calling format, so a vendor had to build the same integration several times. With MCP, a vendor builds one server and any compatible assistant can use it. The protocol defines three primitives: resources (data the assistant can read), tools (functions the assistant can call) and prompts (templated workflows).

Anthropic launched MCP with SDKs for Python and TypeScript and reference servers for tools including Google Drive, Slack, GitHub and Postgres. By December 2025, Anthropic reported more than 97 million monthly SDK downloads across Python and TypeScript.

Which AI companies adopted MCP?

OpenAI, Google DeepMind and Microsoft all adopted MCP within about five months of the launch. The table lists the major adopters and what each announced.

Anthropic also reports that AWS, Cloudflare, Google Cloud and Microsoft Azure now offer deployment support for MCP servers. The Agentic AI Foundation, a directed fund under the Linux Foundation, was co-founded by Anthropic, Block and OpenAI, with support from Google, Microsoft, AWS, Cloudflare and Bloomberg. MCP's governance model stayed the same after the donation, with maintainers continuing to set direction through community input.

Which software vendors publish MCP servers?

Software vendors publish MCP servers so that any compatible assistant can work inside their product. Vendors with official or widely used servers include GitHub, Linear, Notion, Slack, Atlassian, Stripe, Sentry, Asana and Cloudflare. AI coding tools were the first category to adopt MCP at scale, because connecting an assistant to a file system, terminal, database and version control maps directly onto the protocol.

For buyers, MCP support is now a practical evaluation question next to security and permissions. A tool without an MCP server depends on custom integrations to work with AI assistants, which are slower to build and harder to maintain.

Which categories of MCP servers exist?

MCP servers fall into categories based on what they connect to. The vendors in the table are examples of tools commonly connected through MCP. They are not rankings, and EPR has not tested every server named. The EPR guide to AI tools for communications teams covers the wider tool category.

What does the Slack MCP server do?

The Slack MCP server connects AI assistants to Slack workspaces so they can read messages, search channels, pull thread context and post updates. Slack, which Salesforce acquired in 2020, maintains the server directly, and workspace administrators control access through OAuth scopes.

In practice, an assistant connected to the server can summarize a week of activity in an engineering channel, find the thread where a pricing decision was made, or post an incident update to a named channel. Microsoft Teams and Discord offer comparable integrations in the communication category.

How is Anthropic Computer Use different from an MCP server?

Anthropic Computer Use is a separate capability from MCP. An MCP server gives a model a structured interface to one tool, while Computer Use lets Claude operate a screen by taking screenshots, moving the cursor, clicking and typing.

The difference decides which one fits a task. MCP works when a tool exposes a server, and Computer Use works on software that only has a visual interface, such as a legacy ERP screen or a procurement form. Computer Use is available through the Anthropic API and Claude products. OpenAI's Operator and the open-source browser-use project are the closest alternatives, and the EPR guide to AI agents covers the agents built on these capabilities.

What are the security risks of connecting MCP servers?

Every connected MCP server extends the access an assistant holds inside a company's systems, so a compromised configuration can expose every platform linked to it. In May 2026, EPR reported on a Claude Code trust flaw and MCP token theft, in which one malicious npm package edited a local configuration file and placed an attacker between Claude Code and the SaaS platforms connected through MCP.

Security teams reviewing MCP servers should check which OAuth scopes each server requests, who maintains the server, and where credentials are stored on the user's machine.

What does MCP mean for communications and marketing teams?

MCP gives communications teams a measurable question to ask about any technology vendor: can an AI assistant use this product directly? Anthropic's decision to publish MCP as an open standard and then donate it to a neutral foundation is a documented case of a company giving up control of a protocol to grow the market around its product. The analysis of how Anthropic does PR covers the communications side of that strategy.

Brands that appear in AI answers depend on what assistants can retrieve, and EPR tracks the mechanics in the Citation Share Index 2026.

What is the MCP adoption timeline?

MCP moved from a single-company release to an industry standard in about 13 months. The table lists the dated milestones.