Key takeaways
- MCP is a common way for AI agents to use tools and read data, so every integration doesn't have to be built from scratch.
- It sits under the neutral Agentic AI Foundation, co-founded by Anthropic, Block and OpenAI, with Google, Microsoft and AWS among its platinum members.
- A standard connection isn't automatically a safe one. Permissions, approvals and logging still need designing.
- Ask any AI vendor or agency whether their agents can work with MCP. It helps you avoid lock-in.
MCP in one paragraph
Think of MCP as a universal connector. Before standards like this, every AI assistant needed its own custom integration for every tool: one for your CRM, another for your calendar, another for your document store. MCP gives AI models a shared way to find out what a tool can do and to call it. Tools expose themselves through an MCP server, and AI applications connect as clients. The Linux Foundation calls it the universal standard protocol for connecting AI models to tools, data and applications.
Who is behind it
On 9 December 2025 the Linux Foundation announced the Agentic AI Foundation (AAIF). Its founding projects are MCP, contributed by Anthropic, goose, contributed by Block, and AGENTS.md, contributed by OpenAI. Amazon Web Services, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI are listed as platinum members.
The foundation says there are more than 10,000 published MCP servers, and that products including Claude, Cursor, Microsoft Copilot, Gemini, VS Code and ChatGPT support MCP. Google’s October 2026 business agents connect to MCP servers too, as we covered in our piece on agentic Gemini for business.
Why a small or mid-sized business should care
- Less lock-in. If your tools and agents speak a shared standard, swapping one of them gets easier.
- Faster integration. Existing MCP servers for common tools can shorten the build, which can bring the cost down. Our AI agent cost guide explains what drives budgets.
- More consistent governance. One pattern for tool access makes it simpler to review, log and restrict what agents can do.
Before and after
| Without a shared standard | With MCP | |
|---|---|---|
| Connecting a new tool | Custom integration for each AI app | One MCP server usable by compatible apps |
| Switching AI vendor | Rebuild integrations | Reuse many existing connections |
| Oversight | Different logging in each integration | A common place to apply permissions and logs |
This table shows what the standard is meant to achieve. Real projects still need testing, and not every tool has a mature server yet.
The risks to manage
A standard makes connecting easy, and that’s exactly why it needs care. An agent with broad access to email, files and a CRM can do real damage if it misreads an instruction or gets tricked by malicious text hidden in a document. Good practice looks like this:
- Approve each server. Only connect tools from sources you trust and have reviewed.
- Give least privilege. Read-only access where you can, and separate credentials for the agent.
- Require human approval for sending messages, payments or deletions.
- Log everything the agent does so someone can review it.
- Keep sensitive data out of scope unless the use case clearly needs it.
How to use this when planning an AI project
You don’t need to understand the protocol to benefit from it. Ask your provider three things: do their agents use open standards like MCP, which of your systems will they connect to, and what happens when something goes wrong? If you’re comparing partners, our guide to choosing an AI agent development company has more questions to ask. And to see how we connect agents to the tools you already use, take a look at our AI automation services.
Frequently asked questions
What does MCP stand for?
Who maintains MCP?
Do I need MCP for my business AI agent?
Is MCP secure?
Sources
- Linux Foundation: Formation of the Agentic AI Foundation (9 Dec 2025)
- CIO Dive: Big tech develops open standards for agentic AI
- TechCrunch: Google brings agentic AI to Gemini, starting with businesses