Before MCP, connecting an AI assistant to your product meant a custom plugin for every assistant. MCP replaces that with one server you write once, and any compatible client can use it.
The three roles
- Tools are actions:
create_task,publish_site,send_invoice. - Resources are things to read: files, records, docs.
- Prompts are reusable templates the user can pick.
A single call, end to end
Why it matters for product builders
If your product has an MCP server, every AI assistant becomes a front-end for it. Users don't need to learn your UI to get value — they ask in plain language, and the assistant does the clicks.
That changes how I think about product design:
- The API is the product surface, so tool names and descriptions deserve the same care as button labels.
- Permissions must be explicit — reading is cheap, writing should be deliberate.
- Small, composable tools beat one mega-endpoint every time.
Getting started
Pick the three actions your users do most. Expose them as tools with crisp descriptions. Connect it to an assistant and watch where it stumbles — that's your roadmap.