Skip to contact
Seamwork AI
← Blog

Does MCP lock you in to one AI company?

No. MCP doesn't lock you in to one AI company, because no single company controls it. On 9 December 2025 Anthropic, the company that created it, gave the Model Context Protocol away to a new foundation under the Linux Foundation, co-founded with OpenAI and Block, and the project's rules now reserve no seats for any company.

If you've been putting off automating anything because you don't want to end up stuck with one AI supplier, that's the part of this subject worth your attention. The technology underneath is fairly dull, and you don't need to understand it in detail.

What is MCP, in short?

MCP, short for Model Context Protocol, is an agreed way for AI to use the software you already pay for, such as your email, accounts package or files. Think of the charging port on a modern phone: once everyone agreed on one shape, any charger worked with any device. We've explained how it works, with a worked example, in what APIs and MCP are. The connection that sits in front of a piece of software is called an MCP server. For the lock-in question, the part that matters is simple: any AI application that speaks MCP can use any MCP server.

If you'd like one of your everyday systems connected to AI this way, we're happy to look at what's possible with the software you already have.

Why the big AI companies all adopted it

Anthropic published MCP and made it freely available on 25 November 2024. What happened next is unusual.

Rather than building their own rival standards, the other large AI companies adopted it. OpenAI added MCP support to its toolkit for developers building AI assistants on 26 March 2025, where the release notes list it as the headline change. Google added it to its own Python toolkit in version 1.16.0 on 19 May 2025. By 10 December 2025 Google was running its own MCP servers for its services, starting with Maps, BigQuery, Compute Engine and Kubernetes Engine.

Then the ownership moved. The Linux Foundation announced the Agentic AI Foundation on 9 December 2025, with three founding projects: MCP from Anthropic, goose from Block, and AGENTS.md from OpenAI. Its platinum members are Amazon Web Services, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI. The gold tier includes Cisco, Docker, IBM, Oracle, Salesforce, SAP, Shopify and Twilio.

You may see January 2026 given as the date of that move. We couldn't find a primary source for it. Anthropic's announcement, the MCP project's own post and the Linux Foundation press release are all dated 9 December 2025. Where other coverage disagrees with three original announcements that agree with each other, we've gone with the originals.

Is it really neutral now?

Handing a project to a foundation can be just for show. In this case there are specifics you can check.

MCP is now formally set up as Model Context Protocol a Series of LF Projects, LLC. Contributions are under the Apache 2.0 licence, which in plain terms means anyone can use and build on the work for free. Its governance document says membership of the group that makes technical decisions is for individuals rather than companies, and that "there are no seats reserved for specific companies". The project's own post says the Linux Foundation "will not dictate the technical direction of MCP". The same kind of arrangement looks after Kubernetes and PyTorch, two widely used pieces of software.

It has carried on that way since. Google's Agent2Agent protocol, a separate standard for how AI assistants talk to each other rather than to your software, joined the same foundation on 17 August 2026.

How widely is MCP used?

At the December 2025 handover, Anthropic reported more than 10,000 active public MCP servers and more than 97 million monthly downloads of its developer toolkits, with Claude itself offering over 75 ready-made connections built on MCP. By the July 2026 specification release, the project reported close to half a billion downloads a month across its main toolkits.

Those numbers show that software developers have taken it up. They don't show that businesses have. A download is a computer fetching some code. Nobody publishes a reliable count of small firms with a working MCP connection, so we can't give you one.

What MCP changes for a small business

The connection outlasts your choice of AI. If an MCP connection is built to sit in front of your accounts package, it isn't tied to Claude, ChatGPT or Gemini. If you change your mind about which AI provider you use in eighteen months, the connection doesn't have to be pulled out and rebuilt. That lowers the cost of picking the wrong supplier, which is an easy mistake to make at the moment.

It also changes what's worth asking for when you have connection work done. A one-off connection built for a single AI product only works with that product. One built with MCP has a standard shape that someone else can pick up and look after later.

What MCP doesn't fix

It's worth being clear about the limits.

MCP can't connect software that won't let anything in. Most business software has an API, a service counter that other programs are allowed to walk up to. An MCP connection needs that counter, or at least a proper way to get data out. If your practice management system has no API, no useful export and a supplier with no interest in adding one, MCP doesn't change your situation. That's a supplier problem rather than an AI problem.

The standard is still changing. The rulebook, called the specification, is versioned by date, and the July 2026 release removed the step where the two sides introduce themselves when they connect, and started phasing out the old way AI applications registered. At the same time it introduced a formal promise of at least twelve months' notice before anything is removed, which is a good sign. The practical point is that connections need looking after. They aren't something you install once and forget.

The official directory of MCP servers is still in preview, and its operators say it may change in ways that break things or have its data reset. And MCP doesn't keep anything safe for you. The specification says outright that MCP "cannot enforce these security principles at the protocol level", and that a server's description of what its tools do should be treated as untrusted unless the server itself is trusted. Deciding what a connection may read, and what it may do, is still a decision for you.

If working out those permissions is the part you'd rather not handle, we're happy to set up the connection with sensible limits and explain what it can and can't touch. You can read more about the kind of AI and automation work we do.

What should you do about MCP?

Nothing urgent. There's no deadline here. It's simply one less reason to wait.

If you were holding off because you worried about being locked in, that worry is smaller than it was, and you can check that for yourself using the links below. When you next speak to your accounting, CRM or booking software supplier, it's worth asking whether they offer an MCP server. The answer tells you a lot about how easy any AI work on top of their software will be. And if you'd like a hand connecting AI to the systems you already use, get in touch and we'll talk it through.

Sources

Checked on 3 November 2026. This subject moves quickly: specification versions, membership lists and adoption figures all change, and the registry is explicitly still in preview. Verify anything you plan to spend money on against the links above rather than against this article.

The next step

The cost of waiting is not zero.

Somewhere in your market, a business the same size as yours has stopped writing follow-ups by hand. They are quoting faster than you, chasing money you are still chasing manually, and starting Monday already knowing where they stand. That gap does not stay the same size.

Half an hour on the phone. You describe your week, we tell you honestly which parts of it are worth automating and which are cheaper left alone. No pitch, and nothing to sign - you decide whether you want to take it further.

Book a free 30-minute call

Or just ring us on 07816 970 768, or email hello@seamwork.co.uk.

We reply within a working day. If we cannot help, we will tell you straight away - no pressure, no commitments.

We explain what we do with this in our privacy notice.