Turn new content into segmented, tested email sequences

Turn new content into segmented, tested email sequences

Splits opted-in HubSpot contacts by one field, writes each segment an email from your asset, A/B tests it and follows up openers.

Set up the tools below, then run the steps in order for the user, carrying each step's results into the next. The run is done when the user has the outcome below.

Outcome

• Your opted-in contacts split into segments, each contact in exactly one, with a count of those left out for a missing value.
• An approved email per segment for this touch, every stat and claim in it traced to your asset.
• This touch queued in Brew, with its send IDs and the date to run the next touch.
• From the second touch on, opens, clicks and bounces by segment and variant, the variant that led, and what the next touch carries forward.

Inputs

Ask the user for these before you start.

• sequence: the touches in order, each with its asset and the email's job, e.g. 1: three stats from https://acme.example/report, no link; 2: drive to https://acme.example/blog/post; 3: link to the full report
• touch: which touch this run sends, e.g. 1
• segment_property: the HubSpot contact property that splits the audience, e.g. industry, persona, lifecycle stage or pain point
• segments: the values of segment_property to target, one email each, e.g. Cereal, Salty snacks, Canned goods
• consent_property: the HubSpot contact property that is true when someone opted in to marketing email, e.g. marketing_opt_in
• test_element: what the two touch-one variants differ on, e.g. subject line, or call to action
• wait_days: how long to wait after a touch before the next one, e.g. 4
• previous_send_ids: for touch 2 or later, the send IDs the last run returned, with the segment and variant of each
• max_emails: the most email designs to generate in one run, e.g. 8

Set up

Scrape a URL (Firecrawl, tool:firecrawl/scrape-url)

Use the first option your agent supports.

MCP (official, remote)

Add this server to your agent's MCP settings, then sign in when asked.

{ "mcpServers": { "firecrawl": { "url": "https://mcp.firecrawl.dev/v2/mcp-oauth" } } }

Call the MCP tool firecrawl_scrape.

CLI (official)

Install the command, then confirm it runs.

npm install -g firecrawl-cli
firecrawl --version

Run firecrawl scrape.

Set $FIRECRAWL_API_KEY in your environment first (get a key: https://www.firecrawl.dev/app/api-keys).

Brew (tool:brew/get-send, tool:brew/create-audience-from-events, tool:brew/generate-email, tool:brew/send-email)

Use the first option your agent supports.

Notes:

• Get a send: Use the sendId returned by send-email. MCP takes report:{kind:"sends",sendId}. Queued is not delivered; inspect delivery accounting and recipient events before recording delivery. Stop polling when terminal or your execution window ends.
• Create an audience from email events: Builds in the background: poll the audience with include=build until it is ready before sending to it.
• Generate an email: Costs credits and requires a ready brand. If generating, poll get_email. Keep emailId and emailVersionId; pass both to a send or automation to pin the design.
• Send an email: Supply subject and existing subscribed recipients. Use idempotency_key (MCP) or Idempotency-Key (HTTP), reusing it for retries. On confirmation_required, obtain approval and repeat with confirmed:true and the returned confirmation_id. Queued is not delivered.

MCP (official, remote)

Add this server to your agent's MCP settings, then sign in when asked.

{ "mcpServers": { "brew": { "url": "https://brew.new/api/mcp" } } }

• Get a send: call the MCP tool get_email_analytics
• Create an audience from email events: call the MCP tool create_audience_from_events
• Generate an email: call the MCP tool create_email
• Send an email: call the MCP tool send_email

CLI (official)

Install the command, then confirm it runs.

brew install getbrew/tap/brew-cli
brew-cli --version

• Get a send: run brew-cli sends get
• Create an audience from email events: run brew-cli audiences from-events
• Generate an email: run brew-cli emails generate
• Send an email: run brew-cli emails send

Set $BREW_API_KEY in your environment first (get a key: https://brew.new/settings/api).

Search CRM records (HubSpot, tool:hubspot/search-crm-records)

Use the first option your agent supports.

Note: The MCP tool takes up to five groups of six filters and returns up to 200 records per page.

MCP (official, remote)

Add this server to your agent's MCP settings, then sign in when asked.

{ "mcpServers": { "hubspot": { "url": "https://mcp.hubspot.com" } } }

Call the MCP tool search_crm_objects.

API (official)

• Base URL: https://api.hubapi.com
• Endpoint: POST /crm/objects/2026-09/{objectType}/search
• Auth: send the header Authorization: Bearer $HUBSPOT_API_KEY

Note: Use a service key or an app's static access token with the CRM scopes the calls need; paths carry a dated version such as 2026-09.

Before step 1, confirm access to each service with its cheapest read-only call, like a list or a search. Never send or change anything to test access.

Steps

1. Read the asset with Scrape a URL (Firecrawl). Fetch the asset sequence names for touch as markdown. Keep its URL, title, and every stat and claim the email may use, each with the sentence it came from.
2. Check the last touch with Get a send (Brew). Skip on touch 1. Read each send in previous_send_ids; if any went out fewer than wait_days ago, stop and tell the user when to run again. Keep opens, clicks and bounces by segment and variant, and the variant that led in each segment.
3. Pull the segments with Search CRM records (HubSpot). Touch 1 only. For each value in segments, search contacts whose segment_property equals it, with consent_property true and hs_email_optout not set. Count opted-in contacts with no value for segment_property; if they outnumber the segmented ones, stop and tell the user to clean the field first. Put any contact who matches twice in one segment only. Keep each segment's emails.
4. Find who opened with Create an audience from email events (Brew). Touch 2 or later. For each send in previous_send_ids, build an audience of its recipients who opened it, leaving out bounces and unsubscribes, and wait until it has built. Keep each audience ID and size by segment.
5. Plan the angles yourself. For each segment, write how this touch's asset matters to that segment, using only the facts kept in step 1, and the job sequence gives this touch. On touch 1, write two variants per segment that differ only in test_element; on later touches, write one that follows the variant that led in step 2. Show the plan to the user. Keep the approved angle and subject line for each segment and variant.
6. Write the emails with Generate an email (Brew). Generate one email per approved angle, no more than max_emails in all. Check every number and claim against step 1 and flag any that is not in the asset. Show the drafts to the user. Keep each approved email's emailId, emailVersionId and subject, with its segment and variant.
7. Send with Send an email (Brew). Match the consent-qualified addresses to existing subscribed Brew contacts first; stop for the user to prepare any missing contacts. On touch 1, split each segment's contacts alternately into two halves and send each half one variant, 50 inline recipients per send; on later touches, send each segment's email to its audiences from step 4. Give each send its own idempotency key. Keep every sendId with its segment, variant and touch.
8. Confirm the queue with Get a send (Brew). Read each sendId from step 7 until it is terminal or the execution window ends. Keep its status and delivery accounting; a queued send has not been delivered.
9. Summarize the run yourself. Show a table of this touch's sends by segment and variant with their status, and, from step 2, the previous touch's results with the variant that led and what this touch carried forward. Give the sendIds to pass as previous_send_ids and the date, wait_days from today, to run the next touch.

Notes

Running the sequence

Run once per touch: touch 1 sends to every segment, and each later run goes only to people who opened the touch before it, so a sequence of three takes three runs. Touch 1 is the A/B test; later touches send to a saved audience, which the send call can't split, so they carry forward the variant that led instead of testing again. Run touch 1 whenever you publish an asset worth distributing, or monthly with your newsletter links as the asset.

Before the first run

• Segments are only as good as segment_property. Fill it in across your CRM before you rely on it; step 3 stops when most contacts have no value.
• Contacts must already exist in Brew as subscribed contacts, for example synced from HubSpot. Step 7 stops when they don't.
• Read every draft before it sends. Step 6 flags any stat that is not in the asset, so copy stays inside what you published.

Where this comes from

Adapted from how Harmonya's one-person marketing team replaced a single monthly newsletter to the whole list with segmented sequences, as described in this article (sponsored by Brew). The article reports that open and click rates more than doubled against the earlier newsletter, without a baseline or figures; treat that as a reported claim, not a benchmark for this workflow.

Rules

• Use only the services set up above. The read-only calls they need, like listing ids or polling for results, are fine.
• Ask the user before anything that sends messages, costs money, or changes data, and say how many records it touches. One approval covers a batch the user has seen.
• Never print API keys.