# Send & Receive Emails with Claude Cowork (/integrations/frameworks/claude-cowork)

<!-- agent-signals: reading_time_min: 3 · est_tokens: 1007 · updated: 2026-09-06 -->
Related: [Send & Receive Emails with Claude Code](/integrations/frameworks/claude-code.md), [Send & Receive Emails with OpenAI Codex](/integrations/frameworks/codex.md), [Send & Receive Emails with OpenClaw](/integrations/frameworks/openclaw.md), [Send & Receive Emails with Grok](/integrations/frameworks/grok.md), [Send & Receive Emails with Manus](/integrations/frameworks/manus.md), [Send & Receive Emails with Cursor](/integrations/frameworks/cursor.md)



# Give Claude Cowork an email inbox with the AgentMail connector

The AgentMail connector gives Claude Cowork email inboxes it can send, receive, and triage from, authenticated by an AgentMail browser sign-in instead of an API key. Add it from the Claude connector directory, authorize once, and use the AgentMail tools from any Cowork session.

## Do this

1. Get Claude Desktop with the Cowork tab and an AgentMail account from `https://console.agentmail.to`. No environment variable or key is needed.
2. Open Customize in the Claude Desktop sidebar, go to Connectors, and search for `AgentMail`.
3. On the verified AgentMail card ("Email inboxes for AI agents"), select the plus button.
4. A browser tab opens the AgentMail sign-in. Sign in and click Allow to grant Claude access. The connector then shows a connected checkmark.
5. Switch the composer to Cowork and prompt:

```text
Create an AgentMail inbox and send a hi email from it to you@example.com.
```

6. On the first AgentMail tool call, Cowork asks to use the tool. Choose Allow once, or Always allow to stop the prompts for that tool. Cowork creates the inbox, sends, and reports the `@agentmail.to` address.
7. Reply to the test email from your own account, then ask Cowork to check the inbox and respond. It reads the thread and replies in the same conversation.

## Facts

* The connector is in the Claude connector directory as AgentMail (verified, "Email inboxes for AI agents"). A separate community connector, AgentMailer, is not the same integration.
* Authentication is an AgentMail browser sign-in and OAuth Allow when connecting. No API key, no environment variable.
* The connector loads when a session starts. A session started before connecting has no AgentMail tools; start a new one.
* Cowork requests approval on each AgentMail tool call. Allow once approves a single call; Always allow stops the prompts for that tool.
* Because every tool call is gated by approval, a request embedded in an inbound email cannot act on its own.
* Creating an inbox without a requested name lets AgentMail generate an available username. A taken username fails with `403` listing up to 3 available alternatives in `suggestions`.
* The AgentMail plugin (from the plugin marketplace) is an alternative that adds email skills on top of the same hosted tools. See [MCP and Skills](/integrations/mcp-and-skills).

## Not supported

* A session started before the connector was added has no AgentMail tools. Start a new session.
* A request embedded in inbound email is not executed directly. Cowork still asks for tool approval.
* There is no API key or environment variable to set. Authorization happens in the browser when you connect.

## Errors

| Error            | Status | Cause                                                                    | Fix                                                                  |
| ---------------- | ------ | ------------------------------------------------------------------------ | -------------------------------------------------------------------- |
| `resource_taken` | 403    | The requested inbox username belongs to another organization.            | Retry with one of the up to 3 alternatives in `suggestions`.         |
| `already_exists` | 403    | An inbox with the requested username already exists in the organization. | Retry with a username from `suggestions`, or use the existing inbox. |

## Verify

In a new Cowork session with the connector added, prompt `List my AgentMail inboxes`. Success is Cowork calling the AgentMail list-inboxes tool and answering with the inbox addresses. An empty list is a valid answer.

## Related

* [MCP and Skills](/integrations/mcp-and-skills): the hosted tool catalog and the AgentMail plugin that adds email skills.
* [Conversations](/core/conversations): the API behind triage: list threads, read whole conversations, and track their state with labels.
* [Agent safety](/advanced/safety): the untrusted-inbound model and how to apply it to agents you build.