The Model Context Protocol (MCP), the standard that lets an AI agent like Claude or ChatGPT connect to your business tools, just received its largest update since launch. Published on July 28, 2026, this revision changes how AI agents talk to your software, with real consequences for the reliability and security of connections that more and more small businesses use every day.

In brief

• On July 28, 2026, the Agentic AI Foundation (a Linux Foundation body) published the largest revision of the MCP protocol since Anthropic launched it in November 2024 (source: official Model Context Protocol blog).
• MCP is the open protocol that connects AI agents (Claude, ChatGPT, Gemini, Copilot) to business tools: CRMs, email, databases, and automation platforms such as n8n or Make.
• The protocol becomes "stateless": connections are more stable and easier for software vendors to deploy at scale.
• Security is strengthened: new server identity verification (RFC 9207) and a replacement for the old application registration system.
• Some older features are being retired, with a 12-month transition window for vendors that relied on them.
• According to the foundation, MCP development kits total nearly 500 million downloads per month, evidence of the protocol's massive adoption across the software industry.

What MCP is, and why a small business should care

MCP is an open protocol, a kind of common language, that lets an AI agent fetch information or trigger an action in a third-party application without every software vendor having to build a custom integration for each AI. In practice, when a business owner connects Claude or ChatGPT to their email, their CRM, or an automation tool like n8n, that connection very often runs on MCP, even though the end user never sees its name.

The protocol was created by Anthropic in November 2024, then donated in December 2025 to the Agentic AI Foundation, a neutral body hosted by the Linux Foundation and co-founded with OpenAI and Block, with support from Google, Microsoft, AWS, Cloudflare, and Bloomberg. This shared governance matters: MCP does not belong to a single AI vendor, which is why Claude, ChatGPT, Gemini, and Microsoft Copilot all support it.

What actually changes with the July 28, 2026 update

Three older protocol features (named Roots, Sampling, and Logging in the technical documentation), along with the legacy HTTP+SSE connection mode, are now marked deprecated, with a minimum 12-month window before final removal. Vendors that relied on them have time to migrate.

Why this update matters for a small business

Most small business owners will never need to configure MCP directly: it is an invisible technical layer managed by vendors (Anthropic, OpenAI, Google, Microsoft) and by automation platforms (n8n, Make, Zapier). Still, three practical consequences are worth knowing:

1. More reliable connections. An integration between an AI agent and a business tool (CRM, email, shared spreadsheet) should see fewer disruptions, particularly during usage spikes.
2. Stronger security by default. The new identity verification rules reduce the risk of a malicious server impersonating a legitimate tool, a sensitive point for any company that connects an AI agent to customer data.
3. A risk of breakage for custom integrations. A small business that had a custom MCP connection built (for example by a contractor, to link an AI agent to an internal system) should check with that contractor that migration to the new version is planned within the 12-month window.

What a small business can check right now

Limitations to keep in mind

This update does not change what an AI agent can do, it changes how it technically communicates with tools. A small business should not expect new user-facing features in the short term.

The 12-month window is not unlimited. Integrations built on the deprecated features will eventually stop working. A custom integration project started today should account for this constraint in its specifications.

Neutral governance does not remove implementation differences. Even though MCP is a shared standard, each vendor (Anthropic, OpenAI, Google) can adopt it at its own pace. A small business that depends on a specific integration should verify the compatibility announced by its provider rather than assume it.

FAQ

What is the Model Context Protocol (MCP) in one sentence?

MCP is an open protocol, created by Anthropic in 2024 and managed since December 2025 by the Agentic AI Foundation (a Linux Foundation body), that lets an AI agent connect to external business tools (CRM, email, databases) following rules shared by every vendor.

Does a small business need to act now because of this update?

In most cases, no. Small businesses using official connectors (Claude, ChatGPT, Copilot, or ready-made connectors on n8n and Make) have nothing to do: vendors manage the migration. Action is only needed for custom-built MCP integrations.

Is MCP controlled by a single company?

No. Anthropic created MCP, then handed it in December 2025 to the Agentic AI Foundation, a neutral Linux Foundation body co-founded with OpenAI and Block, with support from Google, Microsoft, AWS, Cloudflare, and Bloomberg (source: Linux Foundation, December 2025 announcement).

Which AI tools already use MCP?

Claude, ChatGPT, Gemini, Microsoft Copilot, and many automation platforms such as n8n support MCP to connect their agents to external tools (source: Agentic AI Foundation).

Going further

This update confirms that connections between AI agents and business tools are becoming more robust, a positive signal for any small business considering automating part of its operations with AI. To dig deeper into the protocol's basics, read our guide Understanding the Model Context Protocol, our article on securing AI agent connectors, or see how other small businesses built their own automation projects in our customer success stories.