Build with AI

Courier offers four ways to connect an AI agent. Pick by how you want the agent to work.

MCP server

Structured tool calls inside an agent session.

Courier CLI

Shell access for scripting and CI/CD.

Courier Skills

Best-practice packs that guide your assistant.

Docs for agents

Machine-readable indexes agents can fetch.

Which one to use

Reach for	When
	Your agent works in a session and should discover Courier’s tools on its own, with typed parameters instead of shell commands.
	You are scripting, running in CI, or want the agent to compose shell commands and read `--format json` output.
	The agent already has API access and needs to know the right way to use it: which endpoint, which shape, which conventions.

They combine. A common setup is the MCP server for tool calls plus Skills for the conventions, with the CLI reserved for CI.

Prerequisite: create an API key

All four need a Courier API key. , then replace YOUR_COURIER_API_KEY wherever the setup pages ask for it. Keys belong to a single . Use a Test key while you build. Every new workspace has a Test environment with a testing Provider, so your agent can send before you configure anything.

Docs for agents

Give your agent the docs in a machine-readable form:

URL	Description
`llms.txt`	Index of every documentation page
`llms-full.txt`	Full documentation in a single file

Both follow the llms.txt standard and are generated from these docs.

⌘I

Get started

Tools

MCP server

Courier CLI

Courier Skills

Docs for agents

Which one to use

Prerequisite: create an API key

Docs for agents

MCP server

Courier CLI

Courier Skills

Docs for agents

​Which one to use

​Prerequisite: create an API key

​Docs for agents

Which one to use

Prerequisite: create an API key

Docs for agents