Introduction
Sales follow‑ups are where deals live or die. In 2026 you can build a ChatGPT plugin with little or no code that reads a CRM record or meeting notes, drafts and sends personalized follow‑ups, schedules reminders, and logs activity — all inside ChatGPT or a shared workspace. This guide walks you through a practical, no‑code path: design the workflow, host lightweight endpoints with a visual tool, package a skills‑first plugin (or use ChatGPT’s plugin‑creator), test locally, and publish to the ChatGPT plugin directory for teammates or public users. Wherever I reference platform behavior or the OpenAI publishing flow I cite official docs so you can follow exact steps. (developers.openai.com)

Quick architecture overview

A ChatGPT plugin that automates follow‑ups usually has three parts: - Trigger: a user prompt or a skill that recognizes “send a follow‑up” in conversation.
- Action endpoints: hosted webhooks/APIs that perform tasks — create CRM tasks, send email, schedule meetings, or update records.
- Plugin manifest / package: metadata and (where needed) an OpenAPI/MCP description that tells ChatGPT how to call your endpoints or skills.

OpenAI’s current builder model supports both skills‑only plugins (simple, local capabilities) and packages that include server endpoints (MCP/Apps) for external actions; you can create a skills‑first plugin without writing server code and add hosted webhooks later. (github.com)

Step 1 — Design the follow‑up workflow (15–30 minutes)

Decide the exact outcome and data flow. Example minimum for a “post‑meeting follow‑up”: - Input: meeting notes or deal record (name, company, last touch date, next steps).
- Output actions: (a) draft personalized follow‑up email, (b) send email via SendGrid (or create a HubSpot marketing/email activity), (c) create a CRM task or update deal stage, (d) schedule a reminder in calendar or task queue.

Write simple instructions for each action (what fields the endpoint needs, who the email should come from, whether a human should approve the draft). This clarity saves iteration when wiring no‑code tools.

Step 2 — Pick no‑code hosting & connectors

You’ll need a place to receive ChatGPT calls and run integrations. No‑code platforms that expose HTTPS webhooks or API endpoints work well: - Pipedream: create HTTP sources (endpoints) and chain actions (transform, call APIs). Good for quick REST calls and templates. (docs.automationanywhere.com)
- n8n: visual workflow builder with webhook nodes, API nodes, and credential stores; self‑host or use cloud plans. (n8n.io)
- Bubble: API Workflows let you expose endpoints and trigger backend workflows (useful if you already run your CRM/prospect DB in Bubble). (scribd.com)

These platforms let you authenticate to SendGrid, HubSpot, Gmail, Calendly, or other services via stored credentials so your plugin endpoint can act without manual coding.

Step 3 — Build the endpoints (no code)

Example sequence in Pipedream or n8n: 1. Create an HTTP webhook that accepts a JSON payload (contact, message prompt, follow‑up template).
2. Add a transformation step that fills a prompt template (e.g., "Write a 3‑sentence follow‑up referencing X and proposing a next call").
3. Call the email provider (SendGrid API) or HubSpot Contacts/Engagements API to send or log the message.
4. Create a CRM task via HubSpot API or your CRM’s REST endpoint.
5. Return a concise JSON result (status, message id, CRM link) to ChatGPT.

Use each platform’s credential manager to store API keys (so keys never appear in chat logs). Platform docs show how to create HTTP/webhook endpoints and add API actions. (docs.automationanywhere.com)

Step 4 — Create the plugin package (skills or MCP)

No‑code options: - Skills‑first (recommended for minimal friction): author one or more SKILL.md files in a plugin folder that describe the follow‑up capability and when to invoke it. Skills let ChatGPT run the workflow logic inside conversations without a hosted API. Use ChatGPT’s @plugin‑creator to scaffold a plugin package automatically. (github.com)
- Hosted endpoints / MCP: if your endpoint runs externally, reference it in the package (OpenAI's packaging expects a plugin.json / .codex‑plugin manifest and, when applicable, an OpenAPI or MCP server description). Package docs explain how to include MCP connections and required metadata (privacy URL, support URL, icons). (developers.openai.com)

If you prefer zero ZIPs and manual files, the @plugin‑creator skill in ChatGPT Work mode can create the manifest and skill files for you; then point the package at your webhook URL.

Step 5 — Local testing and approval flow

• Install the plugin into a personal marketplace or workspace (OpenAI supports testing packages locally before public submission). Use test conversations to trigger the skill and confirm the webhook call sequence. (github.com)
• For sending real emails, enable sandbox modes (SendGrid) or send to a test mailbox to verify DNS/SPF/DKIM and templates. SendGrid and most email providers document sandbox/testing guides. (twilio.com)

Step 6 — Publish: verification, packaging, and submission

When you’re ready to share: - Prepare metadata: privacy policy URL, support URL, logo, and clear description of capabilities and data use. OpenAI’s plugin submission flow requires these fields and may require organization/business verification. (developers.openai.com)
- Submit the packaged ZIP (skills, .codex‑plugin/plugin.json, optional MCP descriptors) via the developer portal. After approval you can choose when to publish to the universal directory. Automated checks will scan hosted MCP servers after publication. (developers.openai.com)

Note: older ai‑plugin.json well‑known patterns are legacy; follow current packaging and MCP guidance in the OpenAI docs when preparing your package. (isitready.dev)

Step 7 — Production best practices

• Security: store API keys in the no‑code platform secret manager; require OAuth where supported; never hardcode secrets in the manifest. (docs.automationanywhere.com)
• Rate limits & retries: add exponential backoff for API calls and return clear error payloads to ChatGPT so the assistant can explain failures.
• Data minimization: send only the data needed for a follow‑up draft; show users what will be sent and require explicit approval for high‑risk actions (sending emails, changing deal stages). OpenAI guidelines require clear user consent for plugin actions. (developers.openai.com)

Conclusion By combining ChatGPT’s skills and plugin packaging with no‑code webhooks in Pipedream, n8n, Bubble or similar platforms, you can create a practical sales follow‑up automation without heavy engineering. Start with a skills‑first prototype (use @plugin‑creator to scaffold), wire a hosted webhook for actions you can’t run inside the assistant, test thoroughly with sandbox email/CRM workflows, and follow OpenAI’s packaging and submission steps to publish. With a clear workflow, safe credentials, and careful user confirmation, a no‑code ChatGPT plugin can save reps hours and keep more deals moving forward. (github.com)

Further reading and references - OpenAI: Upload and submit your plugin; packaging and guidelines. (developers.openai.com)
- ChatGPT Learn: Build plugins and @plugin‑creator quickstart. (learn.chatgpt.com)
- No‑code endpoint builders: Pipedream HTTP sources, n8n webhooks, Bubble API workflows. (docs.automationanywhere.com)
- Email delivery: SendGrid Email API and sandbox testing. (twilio.com)

If you’d like, I can: (a) generate the exact SKILL.md text for a “meeting follow‑up” skill you can paste into a plugin package, or (b) produce a step‑by‑step Pipedream workflow (with node names and sample JSON) that sends a follow‑up email and logs a HubSpot task. Which would help most?