Product 1 min read
MCP for your agent, REST for your scripts, one key
Every Cotlet integration is built as an API first, with an MCP server on top. Why we ship both, and how one key and one URL cover every app you connect.
Cotlet team
Most people meet Cotlet through their agent: Claude Code, Codex, Cursor, ChatGPT or n8n. But plenty of the work you want done is not a conversation. It is a cron job, a script, a step in a pipeline. So every integration ships two ways.
API first, MCP on top
We build each integration as an API first. The MCP server is a layer on top of it. That order is simpler to build and to test, and it means you get both at no extra cost:
- MCP for agents. Your agent lists the tools, reads their schemas and picks one.
- REST for scripts. The same actions as plain HTTP endpoints.
Inputs and outputs share one JSON Schema, so a tool and its endpoint always take and return the same shape.
One server, every app
You do not add one MCP server per app. You add Cotlet once:
claude mcp add --transport http cotlet https://cotlet.dev/mcp/v1 \
--header "Authorization: Bearer YOUR_COTLET_KEY"Every app you connect shows up as more tools on that same server. Connect a new one later, and your agent sees it on the next tool list.
One key
The same key works for MCP and REST. Revoke it, and both stop.
Agents can check before you sign up
Without any key, an agent can search the catalog, read an integration’s tools and their status, and request a missing app. It cannot run an action: that needs your account and your sign-in. If it tries, it gets a clear error with a sign-up link instead of a silent failure.
Setup snippets for each client are in the API docs.