What Is MCP? (Model Context Protocol Explained)

Updated July 23, 2026 · Data as of 2026-07-09

MCP, the Model Context Protocol, is an open standard for connecting AI assistants to outside systems such as databases, file stores, ticket trackers, and internal APIs, so that any compliant assistant can use any compliant connector.

Before a standard existed, every integration was bespoke. If you wanted an assistant to read your issue tracker, someone had to build that specific bridge for that specific assistant, and the work did not carry over to the next tool you tried. MCP replaces that with a common interface. A connector, usually called a server, describes what it can do. The assistant discovers those capabilities and calls them when relevant. Swap the assistant and the connectors keep working.

In practice this shows up as a list of available tools inside your assistant. A file server lets it read a directory. A database server lets it run queries. A browser server lets it fetch a page. The model does not gain new abilities from this; it gains reach. The reasoning is unchanged, but the assistant can now act on real data instead of asking you to paste it.

The security surface is worth stating plainly. A connector is code running with whatever access you granted it, and the assistant decides when to call it based on text it has read. If that text came from an untrusted source, a web page, an email, a pull request comment, it can attempt to steer those calls. Grant the narrowest access that does the job, prefer read-only where possible, and be deliberate about which connectors are enabled at the same time as untrusted input.

Why it matters when you choose a tool

MCP support is a portability signal rather than a feature you use directly. A tool that speaks it can plug into the growing pool of connectors other people maintain, and the setup you build is not stranded if you switch tools next year. If your work involves your own data, internal systems, or a stack of niche services, weigh it heavily. If you use an assistant for self-contained tasks such as drafting or one-off coding questions, it should not drive your decision at all.

Frequently asked questions

Do I need MCP to use an AI assistant?

No. Most people never configure a connector. It matters when you want the assistant to reach systems the vendor did not build an integration for, which is common in company settings and rare for casual personal use.

Is MCP tied to one vendor?

It is an open protocol with implementations across several assistants and a large ecosystem of community connectors. That portability is the main argument for preferring it over a proprietary plugin system.

Is it safe to install connectors?

Treat a connector like any other software you run: check who maintains it, prefer read-only access, and give it the smallest scope that works. The extra risk beyond ordinary software is that the assistant chooses when to call it, sometimes influenced by content it reads.

Back to the glossary · Take the quiz · How we score tools