Draft Your Lifecycle Program in Claude or ChatGPT, With Your Real Business Context (Not a Prompt Library)
You already use Claude or ChatGPT every day, and you have been burned twice by AI in marketing: once by a tool that produced nothing you could use, and once by one that did something you never approved. So when someone suggests “just ask ChatGPT to design the lifecycle program,” you can hear both failure modes at once.
Both are avoidable, and the way to avoid them is the same. The first failure comes from an assistant that does not know your company; the second from one that can act on it without you. The working pattern sits between the two: the assistant reads everything about your marketing, writes only drafts, and a person approves each one in the tool that runs it. This post explains why prompt libraries produce generic programs, compares what the lifecycle vendors’ MCP servers actually let an assistant do as of September 2026, and then walks through one real session, from business context to three approved automations.
Why prompt libraries fail
Search for lifecycle prompts and you find libraries: ten prompts for marketers from a lifecycle vendor, thirty prompts from a prompt academy. They are fine as far as they go. Lean On Marketing’s 2026 guide to AI email prompts makes the honest point that AI output matches a human writer on engagement and beats one on production speed only when the prompt carries the role, the audience, the offer and the voice.
That condition is the whole problem. A prompt carries what you paste into it. So the assistant designing your lifecycle program does not know that the current offer is a 30-day pilot rather than the free audit you retired in July. It does not know your ICP the way your account defines it. It does not know that a proposal-stage nurture is already live with a day-7 check-in, so the program it designs sends a second check-in on day 7. It produces a plausible lifecycle program for a company like yours. It cannot produce a correct one for your company, because it has never seen your company.
This is not a wording problem, and a longer prompt does not fix it, for the reasons in why AI content sounds generic. It is a context problem, and the fix is a connection to where the context lives.
Connectors fix the context problem, and create a second one
The lifecycle vendors have noticed. In the first half of 2026, Klaviyo expanded its MCP server to Claude and Cowork, per its newsroom announcement; Braze launched agentic features around its Agent Console; Customer.io maintains MCP documentation that was updated this month. An assistant connected to one of these can read your data, which solves the first failure.
It also raises the second. Once an assistant can write to a lifecycle tool, the question becomes what a write does. There are three possibilities: the server refuses writes, the server writes straight to the live program, or the server creates a draft that a person publishes. Only the third one is safe to hand to an assistant that is reading web pages and CRM notes all day, and the reasoning for that is in the draft-only rule. So the useful comparison is not “which vendor has an MCP server.” It is what the server does when the assistant asks it to create an automation.
What the lifecycle MCP servers actually let an assistant do
From each vendor’s public documentation, as of September 2026. Where the docs do not say, the table says so.
| Tool | Reads | Can an assistant create an automation or flow? | Live or draft? | Source |
|---|---|---|---|---|
| HubSpot (connector for Claude) | CRM records, campaign and email data | No; the connector cannot create or edit workflows or sequences | Text in the chat to paste by hand | HubSpot docs |
| Braze | Campaign and canvas data | Not as documented; writes cover templates and content blocks | Live for the objects it writes | Braze MCP docs |
| Klaviyo | Account data, campaigns, flows | Flow creation is in the tool list; documented use cases are reports and briefs | Live; the server also includes a campaign send tool | Klaviyo blog and developer docs |
| Ortto | Audience and campaign data | No journey creation tool | Not applicable | Ortto help |
| Customer.io | Customer data, campaigns, templates | Writes exist | Live edits sit behind an admin setting that is off by default | Customer.io docs |
| Marqeable | Business context, campaigns, content, calendar, automations | Yes: automation drafts with steps, timing and exits | Draft; a person approves and publishes | This post |
The pattern is the one that runs through this whole subject. The incumbents lean read-only, which is safe and leaves you pasting. The challengers lean live, which is powerful and exactly what you were burned by the second time. Draft-and-approve is the middle, and as of this writing it is mostly empty. The evaluation checklist for any other tool you are considering is in does it have an MCP server?.
The working pattern
Read everything. Write drafts only. Approve in the tool. In practice that means four steps, and the first two are the ones prompt libraries skip.
- Read the business context. Brand, ICP, voice, current offer, straight from the account, not from memory.
- Read what already runs. Every automation and its plan, plus the calendar, so the new program neither duplicates nor collides.
- Draft each automation with a trigger, steps, per-step timing, quiet hours, exit rules and an approval gate on any step a person would normally send.
- Critique, revise, approve. The assistant revises the draft on request; a person publishes it.
If you have not yet decided which stages the program should cover, do that first with the stage-and-trigger map for sales-led SaaS. The session below assumes the map exists.
One real session
Marqeable’s MCP server is at https://app.marqeable.com/mcp, a remote server over HTTP with OAuth 2.1. You sign in with your normal Marqeable login and the assistant sees what your account sees. In Claude: Settings > Connectors > Add custom connector, paste the address, Connect. In ChatGPT: Settings > Connectors, Developer mode under Advanced, Create with the address and OAuth. The session reads the same in either.
You: Run whoami, then get_business_context, then get_automation_overview. I want to build out the lifecycle program from my map. Tell me what we have first.
Claude: Signed in as the marketing lead on the Northwind account. Business context: B2B SaaS for finance teams; ICP is the first controller at a 50 to 200 person company; voice is direct and specific; current offer is the 30-day pilot. Automations live: a demo-request follow-up with two steps, and a proposal-stage nurture with a day-7 check-in and a day-14 case study, no exit on reply. Calendar for the next four weeks: nurture email on the 4th, webinar reminder on the 16th, LinkedIn post on the 18th.
You: Draft three automations: demo no-show, stalled deals at 14 days, and closed-won welcome. Follow the map. Quiet hours on all of them. Approval gates on every SMS and on the founder’s welcome note.
Claude: Three automation drafts created.
Demo no-show: triggered when a booked demo passes unattended. SMS 30 minutes after, approval gate; email next morning at 8am; exit on rebook.
Stalled deals: triggered at 14 days of no activity on any pre-close deal. Email on day 0 with a new proof point rather than a check-in; SMS on day 7, approval gate; exit on any reply or closed-lost. I checked the calendar and moved nothing, since these fire per contact rather than on a date.
Closed-won welcome: triggered on closed-won. Founder note within 24 hours inside quiet hours, approval gate; what-happens-next email on day 2 with the kickoff date field; internal reminder if no kickoff is booked by day 5; exit on kickoff held.
All three use the pilot offer and your voice. None is live. Links: [demo no-show draft] [stalled draft] [closed-won draft].
You: The stalled email is too long and it opens with “I wanted to check in,” which is the phrase I told you to avoid. Three sentences, lead with the proof point.
Claude: Revised. The stalled automation’s day-0 email is now three sentences opening with the finance-team result, no check-in phrasing. I updated the automation draft with the revision; the earlier version is superseded. Same link.
You: Also fix the existing proposal-stage automation to exit on reply.
Claude: I read the proposal-stage plan and drafted a revision adding an exit on any reply from the account. It is a draft revision; the live automation is unchanged until you approve: [link].
Four drafts, one revision, one fix to an existing program, and the assistant never touched anything live. The stalled email was wrong the first time, which is the point: the review caught it, in the same conversation, before it existed anywhere but a draft.
Ask for the coverage read before the drafting every time. “What do we have” is a two-tool call that takes seconds and prevents the duplicate day-7 check-in that a prompt library would have cheerfully produced. The assistant is a good drafter; it is a better drafter when it has read the account first.
What approval looks like
Every draft in that session is waiting in Marqeable. Opening one shows the trigger, the steps with their timing, the quiet-hours window, the exit rule and the gate, plus each step’s copy. You edit in place or accept, and you publish. From that moment the automation runs on its trigger; before it, nothing does. The human in the loop is not a slogan here; it is the publish button, and the assistant does not have one.
The same pattern works when the program already exists and you suspect gaps. The 60-minute lifecycle audit uses the read tools to fill in a coverage table, then drafts an automation for each empty row. And if your stages live in HubSpot’s lifecycle property, HubSpot lifecycle stages are not lifecycle marketing maps each stage to the automation that should fire, with HubSpot’s own connector reading the stages in the same chat.
Frequently asked questions
Can ChatGPT or Claude build a lifecycle marketing program?
They can design one well when they know your ICP, offer, voice and what already runs, and badly when they only have a prompt. Through an MCP connector to a system that holds those things, the assistant can also draft the automations directly, with timing and quiet hours, for a person to review and publish.
Why do lifecycle prompt libraries produce generic programs?
Because a prompt carries only what you paste into it. The assistant does not know your current offer, the audience as you actually define it, or that a proposal-stage nurture is already live, so it produces a plausible program for a company like yours rather than a correct one for your company. Context, not prompt wording, is the missing input.
Which lifecycle tools let an assistant create automations through MCP?
As of September 2026, most either read only or write straight to live. HubSpot’s Claude connector cannot create or edit workflows; Braze’s server reads and writes templates and content blocks; Klaviyo’s documented uses produce reports and briefs and its server includes a live send tool; Ortto’s server has no journey creation; Customer.io can write but keeps live edits behind an admin setting. Marqeable’s server creates automation drafts that a person approves.
What does a good session look like?
Read business context, read the existing automations, agree on the stages to cover, draft each automation with steps, timing, quiet hours and exit rules, critique and revise the draft, then approve it in the tool. The assistant does the reading and drafting; the person keeps the judgment and the publish button.
The bottom line
A lifecycle program designed from a prompt library is generic because the assistant never met your company. A program pushed live by an assistant is dangerous because nobody met the draft. The pattern that works reads everything from the account, writes only drafts, and hands each one to a person to approve, and as of September 2026 almost no lifecycle tool’s MCP server offers that middle path. Build the map, connect the assistant, ask what already runs, draft, critique, approve. The program is yours; the typing is not.
See draft-and-approve in practice: Marqeable’s agents build and run your campaigns, run behavior-triggered automations across text and email, answer and qualify every visitor with AI website chat, and tie revenue to the exact message with attribution.
Marqeable runs your campaigns, answers every visitor, text, and email in seconds, and turns them into booked jobs and meetings - even at 9pm on a Saturday. We’re in private beta with a small early cohort. Get early access
