Agent quickstart

SendSets is designed to be operated by your coding agent. The web app is the control room for inspecting results, approving work, and fixing anything that needs attention.

The first useful outcome is a set of campaign ideas that use your actual product. Your agent reads the codebase before it configures sending.

Start with one prompt

Paste this into Claude Code, Codex, or Cursor:

Read https://sendsetsapi.com/claude.md and follow the Sendsets skill it links.
Inspect this codebase and explain what the product does, who it serves, what
useful result it can create for one prospect, and which product events exist.
Propose five distinct outbound workflows that deliver a useful result before
asking for a meeting. Include an example email, Sendsets' role, and missing
pieces. Ground existing capabilities in files. Do not change code or configure
a campaign until I choose an idea.
After I choose, install the CLI, sign me in, run diagnostics, and show me the
exact first email. Ask before any paid mailbox order, test send, or launch.

The agent first gives you five product-specific ideas. After you pick one, it installs the sendsets CLI, opens browser approval, and runs diagnostics. You do not need to find or paste an API key.

Choose the useful result

Good workflows create something the prospect can use: a finished report, preloaded workspace, working preview, or specific fix. The agent should show the signal, data source, action in your product, Sendsets email, product event, and follow-up for each idea. It should distinguish working product capabilities from endpoints or events that still need to be built.

Get one sender ready

The agent asks whether to connect a mailbox you already have or provision a new one. Provider approval still happens in your browser when Google or Microsoft requires it.

If you are operating the CLI directly, the equivalent commands are:

curl -fsSL https://sendsetsapi.com/cli.sh | sh
sendsets login --hostname sendsetsapi.com
sendsets doctor
sendsets mailbox add --provider gmail --email you@gmail.com --smtp-pass "$APP_PASSWORD"
sendsets mailbox test <mailbox_id>
sendsets mailbox warmup enable <mailbox_id>

Warmup is a mailbox setting. Enable or pause it from the agent, CLI, or the mailbox detail screen.

Describe the audience

Tell the agent who you want to email. You can paste one email address, provide a CSV, or describe the audience. For the fastest first run, start with one address you are allowed to contact.

The agent creates the prospect, campaign, sender assignment, and sequence. It then shows the exact first email before anything is sent.

sendsets campaign create --name "Developer tool founders" --daily-limit 25 --stop-on-reply
sendsets campaign add-step <campaign_id> --subject "Quick question" --body-plain "Hi {{.FirstName}}, ..."
sendsets campaign set-senders <campaign_id> --input '{"senders":[{"email_account_id":"<mailbox_id>"}]}'
sendsets contact create --email jane@example.com --first-name Jane --campaign <campaign_id>

The CLI commands are available when you need precision, but the agent should choose and run them during a normal first session.

Test one prospect

After you approve the exact message, the agent validates the campaign and runs the single-prospect test.

sendsets campaign validate <campaign_id>
sendsets campaign test-email <campaign_id> --to you@example.com --mailbox <mailbox_id>

The test proves the whole path works: mailbox authentication, rendering, sender selection, and delivery. The agent reports the result and asks whether to launch.

Approve launch

Launching is always a separate approval:

sendsets campaign start <campaign_id>

Once it is live, use the web app to inspect campaign state, mailbox health, replies, and analytics. Use the agent to create or operate the workflow.

Errors

Every API error includes error, message, code, and request_id. Agent-facing CLI errors also include a concrete fix and exit non-zero. Checklist commands such as doctor, mailbox test, and campaign validate print every failed check so the agent can resolve the issue before asking you to continue. See error codes.