What is MCP and what problem does it solve?
Answer
MCP, the Model Context Protocol, is an open standard for connecting agents to external systems. A server exposes tools, and any client speaking the protocol can use them.
The problem it solves is combinatorial. Before it, connecting M agents to N services meant M times N bespoke integrations. With a shared protocol it becomes M clients and N servers, so each side implements once.
The pieces:
- Server: exposes tools, and optionally resources and prompts. Often a local subprocess, sometimes a hosted endpoint.
- Client: the harness, such as Claude Code or Cursor.
- Transport:
stdiofor a local subprocess,httporssefor remote.
At startup the client asks each server what it offers, and those tool definitions go into the model's context. When the model calls one, the client routes the call to the right server and returns the result.
Worth knowing that most servers implement only the tools part. The specification also covers resources, which are readable data, and prompts, which are reusable templates, but adoption of those is thinner.
Common wrong answer
Describing MCP as an Anthropic product. Anthropic created and open sourced it, but it is an open specification with implementations across many vendors and clients.
Another mistake is treating MCP as required for tool use. It is not. Function calling works without it. MCP standardises the integration so tools are portable between agents, which is a different benefit from the ability to call a tool at all.
Likely follow-ups
- What are the transport options?
- What are the security implications?