Playbook · Newsletter

Playbook: Run a weekly Kit newsletter with agents, and keep a human on the send button

Put an opt-in on every post, let an agent draft each issue from what you published, target the right tags, and approve before anything is scheduled or sent.

Teardowns: 3Stack: Claude / Claude Code · Kit API v4 · Kit MCPConfidence: Medium to high
Sid Bharath · Keanan Koppenhaver · Nathan Barry Verified Oct 16, 2026 Markdown copy

Who it’s for: Creators and solo operators with a Kit list who collect subscribers but send irregularly.

Tools these builders use: Claude Code, a /newsletter skill, Kit API v4, a JSON content index (Sid Bharath); Claude Code’s skill-creator, a Python script, Kit API (Keanan Koppenhaver); Claude with Kit’s AI connector, the Kit MCP (Nathan Barry).

The steps

  1. Put an opt-in on every post. Sid’s agent adds a Kit opt-in to each post he publishes, picking from the opt-ins he has already set up in Kit. He wanted it to create a tailored opt-in per post, but by his account the Kit API treats forms as read only, so that part isn’t built. Source: Sid (teardown, original)

  2. Check what the API can do before you plan. Sid’s breakdown of Kit API v4: broadcasts, subscribers, tags, custom fields and webhooks are fully programmable; forms are read only; sequences are “Read + subscribe”; visual automations have no API support. He notes 120 requests per 60 seconds per key. This is his table, not checked against Kit’s docs Unverified. Source: Sid (teardown, original)

  3. Keep the API key out of the repo. Sid generates a v4 key under Settings, then Developer in Kit, stores it in .env as KIT_API_KEY, keeps .env in .gitignore and tests it against the account endpoint. Keanan’s published skill also needs a Kit API key. Sources: Sid (teardown, original); Keanan (teardown, original)

  4. Draft each issue from what you already published. Sid’s skill reads his content index, keeps non-draft posts from the last 7 days and asks what to do if there are none. Keanan’s skill writes the email from his post. Sources: Sid (teardown, original); Keanan (teardown, original)

  5. Write voice rules down. Sid’s rules, because “the default Claude output sounds like… Claude”: 2-4 short paragraphs, 1-2 sentences per post with a link, a quick “hey”, a “Sid” sign-off, no em dashes or hyperbole. Output is simple HTML and Kit applies the template. Source: Sid (teardown, original)

  6. Send to the right segment, not the whole list. Keanan’s script only targets subscribers tagged CC4M and excludes anyone still in the welcome sequence. Nathan had Claude create a tag and apply it to 65 beta readers before emailing just them. Sources: Keanan (teardown, original, repo); Nathan (teardown, original)

  7. Bring stray subscribers into Kit. Nathan’s beta readers were stuck in a tool with no export. He pasted the page into Claude, had it clean the names and emails, and add them to Kit with a tag, all in one conversation through the Kit MCP. Source: Nathan (teardown, original, Kit MCP docs)

  8. Test with a dry run. Keanan’s skill has a dry-run mode so he can test without emailing the list. Source: Keanan (teardown, original)

  9. Approve, then create and schedule the broadcast. Sid’s skill shows the subject and body and asks before creating a draft broadcast in Kit, then asks again before scheduling (“Tuesday morning” defaults to the next Tuesday at 10:00 AM EST). Keanan’s script creates a scheduled broadcast from a subject, HTML and send time. Nathan confirmed Claude’s draft before it sent. Sources: Sid (teardown, original); Keanan (teardown, original); Nathan (teardown, original)

Postby’s suggestion: none of these sources reports subscriber growth, open or click figures. What they report is consistency: Sid went from “maybe 2 newsletters a month on a good month” to every week. Postby would treat a steady send plus an opt-in on every post as the growth lever and measure it yourself before claiming more.

Human approval points

  • Sid: approves the draft before the broadcast is created, and again before scheduling. “I don’t want the skill to create the broadcast without me seeing it first.”
  • Keanan: dry-run mode before a real send. The post doesn’t describe a review step beyond that.
  • Nathan: confirmed the draft before it sent. Kit’s docs say tools that send broadcasts or delete data always need explicit user confirmation, and the help centre says some actions only run when you click Confirm inside Kit.

Common failure points (as the builders report them)

  • Sid: the newsletter itself was the step he kept skipping. “Half the time I just didn’t do it at all.”
  • Sid: the Kit API can’t create forms, sequences or automations (his account), so those stay manual.
  • Keanan: he was doing the email-the-list step inconsistently before the skill.

Caveats carried over

  • Sid: “about two minutes” and the 30+ minutes before are his own estimates. The post ends with an opt-in for a skill bundle. The skill links siddharthbharath.com while the post is on sidbharath.com. No publish date shown. X: Unverified.
  • Keanan: no results stated. X: @KKoppenhaver (checked 2026-10-13).
  • Nathan: Kit’s founder announcing Kit’s own connector, so it’s also a product launch. The Kit MCP (“AI Connectors”) runs on paid Kit plans (Creator and Pro). The 65 is his count; no open or reply figures.

Examples

Get one teardown a week.

Same format, a new operator every issue. Double opt-in.