How we run WebinarStack with AI agents, and how you can too
WebinarStack's AI agents use the same public MCP server customers connect to. The five saved workflows, the calls each one makes, and what stays out of reach.
WebinarStack's agents work through the public MCP server at https://webinarstack.co/api/mcp, the same endpoint a customer connects to, with the same 40 tools and the same guardrails. This page lists the workflows that server ships as saved prompts, the calls each one makes, what an agent cannot do, and how to run them yourself. Our own two main uses are adding timed actions to webinars and analyzing how they perform.
The product's documentation is public at docs.webinarstack.co, and the API and MCP surfaces described below are the ones it covers. We have not published run counts or task logs for our own agents, so this page does not quote any.
Which workflows can an agent run on WebinarStack?
Five, each shipped as a saved prompt on the MCP server. Each prompt names the tools the assistant should call and in what order, which is what makes a vague request repeatable.
| Saved prompt | What it asks the assistant to do | Calls it names |
|---|---|---|
| Create Webinar | Build a draft from a launch brief | create_draft, then update_schedule, update_registration_page_basics, update_offer_replay_basics, setup_status |
| Audit Webinar Setup | Check a webinar for launch readiness | reads the webinar summary, then setup_status |
| Improve Webinar Conversion | Suggest changes using setup and analytics | reads the summary, then analytics_summary and setup_status |
| Explain Webinar Analytics | Describe aggregate results in plain language | analytics_summary only, with no raw event dumps or attendee lists |
| WebinarStack Setup Checklist | Check title, schedule, registration page, video, offer, replay and a test registration | no named calls |
The call lists come from the prompt definitions in the server, as of 15 November 2026. The prompts also tell the assistant to treat anything you wrote, such as a webinar description, as data and never as an instruction.
How do I draft a webinar with an agent?
Connect the server, pick the Create Webinar prompt and give it a launch brief. The assistant creates a draft, sets the schedule, fills in the registration page and offer basics, then checks what is still missing.
If you would rather start from a plain request, the MCP server guide covers creating a key and connecting Claude or another client. The same guide lists the setup_end_to_end workflow tool, which creates a draft with a schedule and optional branding, email rules and test registrations in one call.
The draft is where the agent stops. You publish from the webinar editor in the dashboard, which lists anything still missing first.
How do I audit a webinar before launch?
Run Audit Webinar Setup with the webinar's ID. The assistant reads the webinar summary and then calls setup_status, which reports what is still missing. It is a read-only job, so a key with only the read scopes is enough.
This is the safest workflow to try first. It changes nothing, and the output is a checklist you can act on in the editor.
How do I configure automations with an agent?
Describe the rules in plain language and the assistant calls configure_automation_sequence, which creates up to 20 rules in one all-or-nothing call. Step-by-step prompts are in build a behaviour-triggered email sequence by prompting an AI assistant.
Timed polls and CTAs on the video timeline work the same way, through the interaction tools. That walkthrough is in add timed polls and CTAs to a webinar by asking an AI assistant.
Read what the assistant built before you rely on it. The check for an existing matching rule or interaction is guidance the assistant follows, not a lock the server enforces.
How do I read analytics with an agent?
Run Explain Webinar Analytics, or Improve Webinar Conversion if you want suggestions. Both call analytics_summary, which returns aggregate figures. The explain prompt tells the assistant not to request raw event dumps or attendee lists.
The suggestions are only as good as the numbers behind them. A webinar with a handful of sessions will give a thin summary, and the assistant should say so.
What can an agent not do on WebinarStack?
It cannot publish a webinar, send bulk email, run a database query, or register real attendees. Those are outside the 40 tools by policy.
- Publishing is a dashboard action behind a publish-blockers check. The one tool with publish in its name publishes the registration widget's style, not the webinar.
- Real registrations are not available through MCP. The MCP tools create test registrations only. Moving a real list is a REST job, covered in move your registrant list with the bulk registrations API.
- Video upload is not a tool. The video provider can be set, but nothing uploads a file.
- SMS is a dashboard action and is not available over MCP or REST.
The REST API is a wider surface than the MCP server. It has routes for publishing, archiving and webhook management that no MCP tool matches. Whether to give an agent REST access is your decision, made with a key scope, and we have not described an agent workflow that uses it.
How do I keep an agent run safe?
A key with the narrowest scopes that do the job, and a review step before anything goes live. Keys split read and write across webinars, sessions, registrations, automations and interactions. Every tool call is rate-limited and written to an audit table with the tool name, the key used and the resource IDs touched.
Treat the first run as a test. Use a draft webinar, read-only scopes for an audit, and write scopes only for the step that needs them.
Where do I start?
Create a key, connect an assistant, and run Audit Webinar Setup on a draft. It takes minutes and changes nothing. From there, building an email sequence by prompting an assistant lists every trigger a rule can fire on, WebinarStack vs AEvent compares the native approach with an automation layer on top of Zoom, and the features page and home page cover the rest of the product.
WebinarStack runs pre-recorded and just-in-time webinars, and does not run live events.