Tempx.

Claude Code + Tempx

Claude Code can already click through your staging app. What it usually cannot do is the other half of every auth flow: check the inbox. Pair it with a Tempx inbox once, and email steps become part of the agent's normal tool loop — including a blocking wait that returns the moment a verification mail lands.

Pair in one command

claude mcp add --transport http tempx https://tempx.uk/mcp

On first use the browser opens tempx.uk: approve one inbox, permissions (read/search/wait, plus send if you want outgoing tests) and an expiry. Claude Desktop, ChatGPT connectors and Z.ai pair through the same flow — details in the pairing guide.

Example prompts that just work

With the connection live, plain English is the whole API:

You have a Tempx inbox at agent-inbox-9c4@tempx.uk.
Test the signup flow on https://staging.example.com:
create an account with the inbox address, wait for the
verification email, extract the code, finish the signup,
and report whether the flow completed.
Send a password-reset email for alice@example.com via
send_email, then read the captured message and check that
the reset link is present and the wording says the link
expires in 30 minutes.

The agent uses wait_for_email to block until the mail arrives, search_emails / get_email to dig through older threads, and reports back with what it asserted — so the run log reads like a test report.

Why the send side is safe to hand an agent

Every outgoing path is capture-only: send_email and SMTP credentials created via create_smtp_credential deposit messages into the paired inbox with the original recipients preserved in the headers. The agent can verify a notification's content and structure, and it is structurally impossible for a hallucinated recipient to receive anything.

Housekeeping

Add the inbox, then just ask

One command now; every future email flow test is a prompt.

Connect Claude

Related