ai · mcp · tooling
MCP: the USB-C port for AI tools.
Before USB-C you needed a drawer of chargers. Before MCP you needed a custom integration for every AI app and every tool. The Model Context Protocol is the attempt to make that drawer disappear.
What it actually is
MCP is an open protocol, introduced by Anthropic in late 2024, that lets an AI application (the client) talk to external capabilities (the servers) in one consistent way. A server can expose three kinds of things:
- Tools — actions the model may call (create an issue, run a query).
- Resources — data the model may read (files, records, docs).
- Prompts — reusable templates the user can trigger.
Write a server once and every compatible client can use it. That is the whole pitch, and it is a good one.
The part people skip: security
A plug that connects your model to your tools also connects other people's text to your tools. That changes the threat model.
- Treat third-party servers like third-party code: read it, pin it, sandbox it.
- Give each server the least privilege it needs. Read-only by default.
- Keep a human in the loop for anything destructive or that leaves the machine.
- Never put secrets where a model can echo them back.
The protocol is the easy part. The hard part is deciding what an AI is allowed to touch — and that is a security question, not a plumbing one.
Why I like it
As someone who builds both web apps and breaks them, I like standards that move the ugly part (integration) out of every project and into one well-audited place. Fewer bespoke adapters means fewer places for bugs to hide.