AI agents · Playwright MCP
Let your agent finish the sign-up: read the verification email
With Playwright MCP your agent can fill in a sign-up form, and then it gets stuck: the next step is in an email. Add Zyntra's MCP server and the agent reads that email itself, takes the confirmation link from it and opens it in the same browser. Nobody has to open an inbox and paste a link.
# Claude Code: a browser and an inbox for the agent
claude mcp add playwright npx @playwright/mcp@latest
claude mcp add --transport http zyntra \
https://mcp.zyntra.app/mcp \
--header "X-API-Key: $ZYNTRA_API_KEY"Open https://staging.your-app.com/signup and create an account
with the email [email protected] and any
strong password.
Then get the last email for that address from Zyntra, open the
confirmation link from its "links" field in the same browser,
and tell me what the page says. If no email arrives, say so and
stop; do not retry the sign-up.The commands are for Claude Code. In Cursor or another client, add the same two servers in its MCP settings; the MCP docs have the JSON.
What happens to the email
- Your app sends it through its usual email provider, to an address on zyntramail.com.
- A real mail server receives it and stores it for your team, with its links pulled out of the HTML and text and its attachments kept.
- Your test reads it with one API request. Mail to an address you did not create in the dashboard is deleted after 1 day.
What the agent does
- Opens the sign-up page and fills in the form with Playwright MCP's browser tools.
- Calls
get_last_emailfor the address you gave it. The call waits up to 30 seconds for the email to arrive. - Picks the confirmation link from
linksand opens it in the same browser, so any session the app set during sign-up is still there. - Reports what the page says. Ask it to delete the email afterwards if you want a clean inbox.
Prompts that keep the run honest
- Give the exact address in the prompt, with a new tag for each run, so the agent never reads an older email.
- Say what to do if the email does not come: stop and report, rather than sign up again.
- Ask for the final page text or a screenshot, so you can see the result and not only the agent's summary.
- When a run works, turn it into a regular test: the Playwright guide has the same flow as code.
Limits to know about
- Agents are not deterministic. Use agent runs to explore and check a flow, and code for the test that gates a release.
- Each Zyntra tool call is one API request on your plan.
- Mail to these addresses is kept 1 day.
Questions
- Do I need Claude Code?
- No. Any agent that can use two MCP servers at once works: one for the browser (Playwright MCP) and one for the inbox (Zyntra).
- My app sends a code, not a link. Does this still work?
- Yes. Ask the agent to read the code from the email and type it into the page. The agent reads it; Zyntra only hands over the email.
- How is this different from the Playwright guide?
- There, you write the test and it runs the same way every time. Here, the agent decides each step from your prompt. The first is for CI; this is for trying a flow or checking it while you build.