How is MCP different from an ordinary API?

An API tells a programmer how to call a service. MCP tells a model, at runtime, which tools exist, what arguments they take and what they return, in a format the model can read and choose from. The tool becomes discoverable instead of hardcoded.

Do I need MCP to give a model tools?

No. Every major model API already supports tool definitions you write yourself, and for one app with three tools that is simpler. MCP pays off when the same tool has to work across several assistants, or when users install connectors you did not build.

What are the security risks of connecting MCP servers?

Two. The server itself runs with whatever access you granted it, so an untrusted server is untrusted code. And everything it returns enters the model's context, so a poisoned document can carry instructions the agent then follows. Install servers you trust, and keep irreversible actions behind confirmation.

Plain-language explainer

MCP, the universal connector for AI tools

What is the Model Context Protocol, and why does it matter?

MCP is a shared way for an AI assistant to discover and call tools. Before it, every assistant needed custom code for every tool: N assistants times M tools meant N times M integrations. With MCP, a tool ships one server describing what it can do, and any MCP-speaking assistant can use it. It is a plug standard, not a model and not intelligence. What changes is how many things an assistant can reach without anyone writing glue code first.

Last reviewed July 25, 2026

Reading is the slow way. Start with a free lesson you can operate right now.

Start free: The harness: the loop, made real →

Free, no code, no signup.

Then go deeper: MCP: the universal connector Locked

What people get wrong

MCP makes the model smarter. It does not touch the model at all. It changes what the model can reach.
MCP is an Anthropic product you have to buy. It is an open protocol with open implementations, and other vendors have adopted it.
Connecting a server is harmless. An MCP server is code with access you granted, and anything it returns lands in the model's context, which is where prompt injection starts.

Where you see it in real products

Coding assistants that can read your repository, run tests and open pull requests.
Desktop assistants that reach your files, calendar or notes through a connector you install.
Internal company assistants wired to a ticket system or a database through one server instead of bespoke code.

Common questions

How is MCP different from an ordinary API?: An API tells a programmer how to call a service. MCP tells a model, at runtime, which tools exist, what arguments they take and what they return, in a format the model can read and choose from. The tool becomes discoverable instead of hardcoded.
Do I need MCP to give a model tools?: No. Every major model API already supports tool definitions you write yourself, and for one app with three tools that is simpler. MCP pays off when the same tool has to work across several assistants, or when users install connectors you did not build.
What are the security risks of connecting MCP servers?: Two. The server itself runs with whatever access you granted it, so an untrusted server is untrusted code. And everything it returns enters the model's context, so a poisoned document can carry instructions the agent then follows. Install servers you trust, and keep irreversible actions behind confirmation.

Related explainers

AI agents Context engineering Prompt injection

More in Agents and tools

Part of See How AI Works, a free interactive course, where you learn how modern AI works by operating it, not watching videos.