Before you start. You need a Brew account and an AI client that supports plugins or
remote MCP connections. ChatGPT, Claude, Grok, Cursor, Codex, VS Code, and
most agent tools do. A verified sending
domain is also required if you
plan to send for real.
1. Connect Your Client
Use the setup guide for your client, then complete its read-only brand check: Connect Your MCP Client covers ChatGPT web and other clients. MCP Authentication and Scoping explains brand access and revocation. For a new brand, follow Configure a Brand and Domain before starting this recipe.Approval and Retry Instructions
Give your agent these working instructions after setup:send_email, test sends included, and the other fail-closed tools such as audit_email and create_inbox_placement_test. Safe Retries has the full contract. For marketing consent, see What You Can Send with Brew.
2. Draft the Email
Generation uses Brew credits. After your brand is ready, replace the brackets and request one draft:create_email. If the result is still generating, it should poll get_email with the returned emailId and runId. Calling create_email again starts another billed design.
Open the preview and the Brew link. Check the facts, branding, images, and destination links. If your client doesn’t display a preview, inspect the design in Brew before approving further work.
3. Iterate in the Same Chat
Identify the existing design by its Brew link or email ID:edit_email, create a version, and use credits. If the edit is still running, poll get_email with its returned IDs. You can restore an earlier version if needed.
A rename, inbox subject change, language change, or folder move alone uses the free update_email_metadata tool, which creates no version.
4. QA with Agent Tools
The full pre-send routine is available as tools:- “Preview this across Gmail, Outlook, and iOS dark mode” runs
test_email_rendering, which starts a paid rendering job and returns apreviewId. The agent pollsget_email_renderingwith thatpreviewIdand returns a screenshot per client. - “Audit this before we send it” runs
audit_emailand returns readiness, check coverage, size metrics, and the findings to fix. A partial result is free and does not establish readiness. - “Send me a test” runs
send_emailwithtest: true, which sends a one-off test to your inbox and never touches your audience. Check your inbox after the tool accepts it. - “Check whether this lands in the inbox or spam” runs
create_inbox_placement_test, a seed send across real providers. It returnsqueuedorrunning, so ask the agent to pollget_inbox_placement_resultswith theemailIdandtestIduntil it completes. - “Validate the Newsletter subscribers audience before we send” reads all pages of the audience’s addresses with
search_contacts, then checks them withvalidate_contactsin batches of up to 100. Review the verdicts before sending. Validation doesn’t guarantee delivery. MCP Tools lists limits and costs.
5. Send It
Choose a sendable marketing domain and an opted-in audience, then ask the agent to prepare the send:send_email returns a confirmation request before a production send. Review the recipients, sender, subject, preview, and timing. After approval, the agent repeats the exact call with confirmed: true and the returned confirmation_id.
A brand API key skips Brew’s confirmation gate. Your working instructions still require approval before sending. MCP Authentication and Scoping describes the connection-specific behavior.
Check the send’s status and your test inbox. An accepted send isn’t proof of inbox delivery.
6. Where This Goes Next
The same session can run the rest of your email program:- Audiences. “Create an audience of contacts on the Pro plan who joined this quarter” (
create_audience). - Contact imports. “Import this CSV of webinar sign-ups” (
import_contacts_csv, orimport_contactsfor rows the agent already holds). Missing contacts are created and existing ones updated, and the batch appears in Import History. On OAuth and organization connections the import returns a confirmation request first. Check how many contacts it would add as subscribed before you approve it. - Automations. “Build a 3-step welcome flow triggered by user signup” (
save_automation), then test and publish it from chat. - Audience automations. “Run this follow-up sequence for Newsletter subscribers” (the
run_audience_automationworkflow, which ends with therun_automationtool). Review tests and a dry-run recipient count before approving the launch. - Analytics. Next week: “How did the newsletter perform against our last three sends?” (
get_email_analyticswithreport: { kind: "sends" }for the per-send stats, orkind: "overview"for the totals, rates, and timeseries the app shows). - Design import. “Turn this Figma frame into an email” (
import_emailwithsource.kindset tofigma), which converts one frame deterministically and for free. - Images.
add_imagecan host an image from a public URL. A local file usescreate_image_upload, a separate upload request, thenadd_imagewith the returned upload ID. That path needs an agent that can send file bytes; attaching a file in chat alone does not complete the upload. Upload a Local Image covers the transfer. - Send control. A gradual send already in flight can be held and restarted with
pause_email_sendorresume_email_send, passing itssendId.cancel_email_sendpermanently stops remaining deliveries.
Troubleshooting
From an Agent or Code
Before translating this recipe to HTTP or the terminal, check which operations each interface offers. Some MCP tools combine several API operations, and some are MCP-only. MCP Tools lists HTTP equivalents where available. CLI Commands lists terminal operations, and Build with Brew compares the four interfaces.Related
- Connect Your MCP Client: setup per client.
- MCP Workflows: the guided multi-step prompts.
- Use Brew with AI Agents: wiring Brew into an agent that also builds your product.
- Use the CLI from AI Agents: the same work from a coding agent’s terminal.
- QA an Email Before a Big Send: the full pre-send routine step 4 runs.
Need Help?
Our team is ready to support you at every step of your journey with Brew. Choose the option that works best for you:- Self-Service Tools
- Talk to Our Team
Search Documentation
Type in the “Ask any question” search bar at the top left to instantly find relevant documentation pages.
Ask ChatGPT or Claude
Click here to open the Copy page menu at the top of this page, then send the page to ChatGPT, Claude, or your editor.