# Who owns MCP now? Nobody. That is exactly why you can build your AI's memory on it.

Before you build anything important on a protocol, you ask one question. Who controls it?

If the answer is one company, you are renting. They can change the terms, close the spec, or steer it somewhere that suits them and not you. People learned that the hard way with API platforms for the last fifteen years.

So here is the honest answer for MCP, the protocol that lets Claude, ChatGPT, and other assistants read your notes and call your tools.

Nobody owns it. Anthropic created it, then gave it away. In December, Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation. It was co-founded with Block and OpenAI, with support from Google, Microsoft, AWS, Cloudflare, and Bloomberg. The maintainers keep running it as an open, community-driven project, and the rules are set in the open instead of by one company.

Read that list again. Those companies compete with each other on almost everything. They agreed to share this. That is the strongest signal you can get that a standard is going to outlast any one vendor's roadmap.

## Why this is the question that matters

There is a lot of noise about whether MCP is winning, whether it is overhyped, whether something newer will replace it. Most of it misses the practical point.

If you are deciding where to keep the knowledge your AI reads, you are not betting on a hype cycle. You are betting on a foundation. The thing you want to know is whether the protocol underneath your notes is owned by someone who can pull it out from under you.

The answer is now no. MCP is an open specification with open-source SDKs, governed by a neutral foundation. No single company can close it, fork your access away, or steer it alone anymore. The rules are set in the open.

That turns "build on MCP" from a gamble into a reasonable default.

## What vendor-neutral actually buys you

Governance is abstract. Here is what it means at your desk.

- **No lock-in to one assistant.** MCP is read by Claude, ChatGPT, Cursor, Copilot, and a growing list of clients. A knowledge base that speaks MCP is not tied to whichever model you prefer this year. Write your notes once. Any MCP client reads the same ones.
- **No lock-in to one vendor's memory feature.** Every AI company is shipping its own built-in memory. Each one lives inside that product. A protocol-level knowledge base sits outside all of them, so switching assistants does not mean starting from zero.
- **A standard that is still here in two years.** Adoption is not a question mark anymore. By the time of the donation, MCP had passed 97 million monthly SDK downloads and more than 10,000 active servers. A neutral foundation plus that much usage is a far safer bet than tying yourself to one vendor's feature.

The protocol being open is not a detail for engineers to care about. It is the reason your knowledge stays yours.

## Where a knowledge base fits

This is the part we care about, so we will say it plainly.

A knowledge base built on MCP is the durable layer underneath whatever assistant you use. You author notes, folders, and tags in Markdown. The built-in MCP server hands them to Claude or ChatGPT on request. When a new client supports MCP, it reads the same notes with no migration. When you switch assistants, nothing moves. The notes never depended on the assistant in the first place.

That only works because the protocol is open and neutral. If MCP were one company's private channel, your knowledge base would inherit that company's lock-in. It is not, so it does not.

You own the notes. MCP is the open road they travel. No toll booth, no single owner.

## The honest caveat

Open governance does not make every MCP server good, and it does not make MCP the right tool for every job. We have written before about [where MCP fits and where it does not](/blog/is-mcp-dead). Point it at the wrong problem and it disappoints, like any tool.

But the specific worry that it is a single-vendor trap is much smaller now. The protocol belongs to a foundation, not a company. That was the one risk worth losing sleep over, and it is far lower than it was a year ago.

## Build on the open standard

If you want your AI to remember what you tell it, put that knowledge somewhere built on a standard nobody can close.

[Hjarni](https://hjarni.com) is a knowledge base with a built-in MCP server. You write notes. Claude and ChatGPT read them, follow your instructions, and remember what you told them across every conversation. Because it speaks the open protocol, the same notes serve every MCP client you use, today and the ones that ship next year.

New to the protocol? [What is MCP](/docs/what-is-mcp) is the five-minute primer, [the MCP knowledge base](/mcp-knowledge-base) explains the pattern, and [combining multiple MCP servers](/blog/combine-multiple-mcp-servers) shows where memory sits among your other tools. When you are ready, connecting [Claude](/docs/claude) or [ChatGPT](/docs/chatgpt) takes about five minutes.

Who owns MCP? Nobody. Write once. You both remember. That is only a safe promise because the road underneath it is open.