MCP for agents

Your coding agent can write the email. Connect Mailclutch and it can send it, read it back and tell you whether it works — without you leaving the editor or opening a browser.

Install it

In Claude Code:

claude mcp add --transport http mailclutch https://app.mailclutch.com/mcp

In Cursor, .cursor/mcp.json:

{
  "mcpServers": {
    "mailclutch": {
      "url": "https://app.mailclutch.com/mcp"
    }
  }
}

In VS Code, .vscode/mcp.json:

{
  "servers": {
    "mailclutch": {
      "type": "http",
      "url": "https://app.mailclutch.com/mcp"
    }
  }
}

Any other editor takes the same address. No sign-up, no key: the first thing your agent does is open a real inbox.

Then ask for things

You never name a tool, you say what you want:

Point this project at a Mailclutch sandbox.
It opens an inbox and writes the SMTP settings into your configuration.
Run the signup flow and confirm the account.
It waits for the email, reads it back, and uses the link or code inside to finish.
Would this land in spam?
It scores the message out of 100 and names what drags it down.
Give me a link to that receipt for the pull request.
It turns the message into a page anyone can open, no account needed.

What it can reach

MCP tools
ToolWhat it does
create_temp_inboxOpens a test inbox and hands back SMTP settings to point your project at.
send_test_emailDrops a sample message in, to check the inbox works before you touch your code.
wait_for_emailWaits for your app's email to arrive instead of polling for it.
list_emailsShows what has been caught: sender, subject, a line of the body.
get_emailReads one message in full — text, HTML and headers.
extract_links_and_codesPulls the confirmation link or the six-digit code out, ready to use.
check_deliverabilityScores the message out of 100 and says what would send it to spam.
create_share_linkTurns a message into a link anyone can open, for a pull request or a chat.
get_connection_statusSays which inbox the agent is working in, and who it is acting as.
link_accountConnects the editor to your Mailclutch account, with one approval in the browser.
list_inboxesLists your projects, so you can say “use the staging inbox” and be understood. With an account.
get_inbox_credentialsGives the SMTP settings of one of your projects — the ones that don't expire. With an account.
search_emailsSearches everything you have kept, not just what is in front of you. With an account.
create_inboxCreates a new project inbox without leaving the editor. With an account.

Keeping your inboxes

A sandbox lasts 24 hours and holds 50 messages. To keep one inbox per project, with its history, ask your agent to connect your account: one approval in the browser, nothing to restart. A sandbox worth keeping can be claimed as it is.

Before you let an agent read your mail

A captured message is written by its sender, and your agent reads it. Mailclutch hands it over marked as something to report on, not to obey, strips anything executable and never opens its links. Still, treat an agent with an inbox like an agent with a browser.

Your app still sends real email

Your project keeps sending to smtp.mailclutch.com on port 587, as to any mail server — see SMTP setup and the framework guides. In CI, where no agent is watching, the read API does the same work with a token.