Skip to main content
Your AI coding agent needs two things to build notifications with Courier: knowledge of how Courier works, and access to your workspace. Courier Skills gives it the first. The MCP server and the Courier CLI give it the second. Install Courier Skills so the agent knows how to build with Courier. It needs no API key:
Then connect the MCP server so the agent can create templates, send test notifications, and read delivery logs in your workspace. In Claude Code:
MCP server has setup for Claude Code, Cursor, Codex, VS Code, and nine other clients. To set up Skills, the MCP server, and a key with one prompt, follow the AI agent quickstart.

Courier Skills

Courier Skills gives your coding agent Courier’s own guidance: how to call the API, which feature fits each job, and the rules each channel needs. It loads only the files a task needs, so the agent spends fewer tokens searching docs. It can also debug the Courier code you already have. The skill touches nothing in your workspace, so it needs no API key. Courier Skills covers the Claude Code plugin and how to check it loaded.

MCP server

The hosted MCP server at https://mcp.courier.com turns the Courier API into tools your agent calls from a chat session. That makes it easy to test and debug with Courier without leaving your agent: send a test notification, check what was delivered, and read the logs when something fails. Ask in plain language:
Tool calls run with the key you connect. A production key sends to real users, so connect a Test key while you build. MCP tools lists every tool and marks which ones write.

Courier CLI

The CLI covers the same API as the MCP server, as shell commands an agent, script, or CI job can run:
Courier CLI covers flags and output formats. CLI recipes has a command for each common task.

Docs for agents

Every docs page is available in a form agents read cheaply, with no API key: llms-full.txt is over 3 MB, enough to crowd out your code in a coding session. Point the agent at the docs MCP server or single .md pages instead.

Feedback API

Your agent is often the first to hit a Courier bug, confusing behavior, or a missing feature. With the Feedback API, it can report the problem to Courier on its own as it works. Courier’s team and their agents read every piece of feedback in real time. Courier Skills tells your agent when and how to report. You can also file a report yourself through MCP or the API. Feedback API has the fields and limits.

API keys

The MCP server and the CLI need a key. Create one in Platform → API Keys. Each key belongs to one environment. A Test key comes with a testing email provider, so the agent can send before you set up a real one. It delivers only to teammates in your workspace.

FAQ

Courier Skills changes the code your agent writes and touches nothing in your workspace. The MCP server gives the agent tools that call the Courier API, so it can create templates, send notifications, and read delivery logs. Most setups install both.
Courier Skills installs into Claude Code, Cursor, Codex, Windsurf, Cline, and any tool that loads agent skills. The MCP server has setup steps for Claude Code, Cursor, Codex, VS Code, Windsurf, Gemini CLI, Antigravity, Zed, Warp, OpenCode, Grok Build, Devin, and Claude Desktop.
An agent connected to the MCP server or CLI sends through the same API your application uses. A Test key keeps sends in your Test environment. A production key reaches real users.
Courier hosts a docs MCP server at https://www.courier.com/docs/mcp. It needs no API key and is separate from the API MCP server at https://mcp.courier.com.