All articles

Manage your automated webinars from Claude: the WebinarStack MCP server

WebinarStack's remote MCP server lets Claude build and manage automated webinars. What its 40 tools do, how to connect it, and what it is not allowed to touch.

Josh Miller

WebinarStack has a remote MCP server, so an AI assistant like Claude can build and manage your automated webinars by conversation. It exposes 40 tools covering webinar setup, scheduling, behaviour-based automations, timeline polls and CTAs, and analytics. Publishing a webinar, bulk email and database access stay out of the agent's reach by policy.

The count is the number of entries in the server's tool allowlist as of 15 November 2026. It changes as tools ship, so treat the families below as the stable part.

What is MCP, and what does it do for a webinar creator?

MCP, the Model Context Protocol, is a standard that lets an AI assistant call a product's tools directly. Without it you copy settings between a chat window and a dashboard by hand. With it, the assistant reads your webinar's setup and changes it itself, and you review the result.

For WebinarStack that means you can say "set up a just-in-time webinar with this title and a reminder sequence" and get a draft you can open in the editor. The assistant is doing what you would do in the dashboard, through the same checks the dashboard applies.

How do I connect WebinarStack to Claude?

You create an API key in the dashboard, then point the assistant at WebinarStack's hosted endpoint with that key as a bearer token.

  1. Create an MCP key in the dashboard under AI Tools. The plaintext key appears once, when you create it. Saved keys show only a masked hint afterwards, and a masked hint is not a usable credential.
  2. For a remote client such as Claude, set the server URL to https://webinarstack.co/api/mcp and send the key as a bearer token in the Authorization header. ChatGPT has its own endpoint at /api/chatgpt/mcp.
  3. For a client that launches servers as local commands, such as Cursor, run npx -y @webinarstack/mcp --server-url https://webinarstack.co/api/mcp. The package forwards to the same hosted endpoint.
  4. Store the key in your client's configuration as WEBINARSTACK_API_KEY.

Keep the key out of everywhere else. Never paste it into a prompt, a screenshot, a support ticket, a source file or a repository. The dashboard's setup page generates the exact configuration for each client, and it is the place to copy from.

Keys are scoped. The scope picker splits read and write access across webinars, sessions, registrations, automations and interactions, and a key can be limited to the MCP surface or the REST API. If you want an assistant that can look but not touch, create a key with the read scopes only.

The server also publishes OAuth discovery metadata. The path the product documents and the one this page describes is the API key.

What can the assistant do?

It can do the setup work of a webinar, the automation rules that make it behave, the timed interactions on the video, and the reporting afterwards. The 40 tools fall into five groups, and the server ships saved prompts and reference resources alongside them.

Webinar setup and reading. Fourteen tools list your webinars, summarise one, create a draft, and update its core details, branding, video provider, schedule, registration page copy, offer and replay settings and replay defaults. It also includes setup_status, which reports what is still missing from a webinar, upcoming_sessions, analytics_summary, and create_test_registration, which lets you walk the attendee path yourself. Test registrations are the only kind the server creates; a real list goes in over the REST API, covered in moving your registrant list. A health tool checks the connection.

Workflows. Five tools do several steps in one call:

  • setup_end_to_end creates a draft with a schedule, and optionally branding, email rules and test registrations.
  • clone_with_new_schedule duplicates a webinar onto a fresh schedule, keeping its branding, content and email, integration, webhook and Meta pixel rules.
  • add_replay_with_new_offer turns on replay and attaches a different offer to the replay timeline.
  • configure_automation_sequence creates up to 20 automation rules in one all-or-nothing call.
  • register_test_attendees registers up to 20 test attendees with behaviour profiles such as attended or no-show.

Automations. Five tools list, read, find matching, update and delete automation rules. Rules can fire on 15 trigger types, including watched_minutes, watched_percent, poll_response, cta_click and branch_taken, and can run four actions: email, webhook, integration and meta_capi. Meta Conversions API rules need the Growth plan and a connected pixel, which the meta_pixels_list tool lists. SMS is a dashboard action and is not available through MCP or the REST API. The product itself has 18 triggers; the three tied to the session type (live_complete, replay_started and replay_complete) are not accepted by any MCP tool that creates a rule, so build those in the dashboard.

Interactions. Six tools list, read, find matching, create, update and delete the polls and CTAs placed on the video timeline. Each one sits at a second-level offset, separately for the live session timeline and the replay. Adding timed polls and CTAs with an AI assistant is the walkthrough, with the prompt to paste.

Widget style and design. Eight tools read and update the embeddable registration widget's style, apply or reset a saved design, switch the colour scheme and set the layout. One publishes the style, which we cover below.

Prompts and resources. The server carries five saved prompts: Create Webinar, Audit Webinar Setup, Improve Webinar Conversion, Explain Webinar Analytics and WebinarStack Setup Checklist. It also carries reference resources: a product guide, authoring guides for automation emails, interactions and Meta pixel events, the OpenAPI spec and a tool catalog. The prompts are the quickest start, because they tell the assistant which tools to call in what order.

The part worth noticing is the write surface. An assistant that can read your webinar stats is useful. One that can build the behaviour-triggered funnel, such as an email when someone has watched 20 minutes or a webhook when they click an offer, is doing the work the product exists for. Building an email sequence by prompting an assistant walks that through end to end, and the watch-time email article and the watch-time webhook article cover what those rules do once they exist.

What stops the assistant from doing damage?

Three things, and one honest caveat.

Dangerous actions are not tools. Publishing a webinar, bulk email, campaign sends and raw database queries are not exposed, and the server's policy treats those terms as forbidden. The assistant builds drafts and you publish. That is deliberate, not a gap.

Two precisions. Two tools can delete: one deletes an automation rule and one deletes an interaction. Both are flagged as destructive, tell the assistant to disable a rule rather than delete it where that would do, and return a snapshot of what was removed. And one tool has publish in its name: widget_style_publish publishes the widget's style settings, and its description says the whole draft style goes live. It does not publish a webinar.

The assistant is told to look before it writes. The find_matching tools check whether a webinar already has a comparable automation rule or interaction, and the create tools tell the assistant to call them first. For interactions, a CTA within ten seconds of the same offset counts as a match, and a poll counts if its headline matches after trimming and lowercasing. This is guidance the assistant follows, not a lock the server enforces, so read what it built.

Calls are recorded. Every tool call is written to an mcp_tool_calls audit table with the tool name, the key used, whether it succeeded, an error category and the resource IDs it touched. Calls are rate-limited too.

The caveat: an assistant is only as careful as its prompt, and customer-written copy such as your webinar description is treated as data. Review the draft before you publish it.

Where is the server listed?

WebinarStack is published to the official MCP registry under the co.webinarstack namespace. The server also serves a server card at /.well-known/mcp/server-card.json that describes the endpoint. If you trust a registry more than a vendor's blog, check the listing there.

What is it good for first?

The fastest uses are the ones that replace repeated clicking. Building a webinar from a launch brief, auditing a webinar's setup before launch, explaining last week's analytics in plain language, and rebuilding a webinar you already run elsewhere are the four we would try first. The last is covered in how to switch webinar platforms, and it is an AI-assisted rebuild: nothing in WebinarStack imports another platform's webinars.

WebinarStack runs pre-recorded and just-in-time webinars, and does not run live events. See what else the platform does on the features page, compare it against an automation layer in WebinarStack vs AEvent, or start from the home page.

Frequently asked questions

Does WebinarStack have an MCP server?
Yes. WebinarStack runs a remote MCP server at https://webinarstack.co/api/mcp, and it is listed in the official MCP registry. An AI assistant that supports MCP, such as Claude, connects with an API key you create in the dashboard and can then build and manage your automated webinars by conversation through 40 tools.
Can the AI assistant publish my webinar?
No. Publishing a webinar is not one of the tools the server exposes, by policy. The assistant creates and edits drafts, and you publish from the webinar editor in the dashboard, which lists anything still missing first. The one tool with publish in its name publishes the style of your embeddable registration widget, not the webinar.
Can I give an assistant read-only access?
Yes. API keys carry scopes, split into read and write for webinars, sessions, registrations, automations and interactions. Create the key with only the read scopes and the assistant can look at your webinars and analytics but cannot change anything.
Can the assistant send email to my registrants or query my database?
No. Bulk email, campaign sends and raw database queries are not exposed as tools. Email goes out only through automation rules that the assistant can create, and those rules fire on the attendee behaviour you define. Every tool call is also recorded in an audit table.
Do I need Claude, or does it work with other assistants?
Any client that speaks MCP can connect. The dashboard's setup page generates configuration for Claude, ChatGPT, Cursor, Windsurf and a generic MCP client. Remote clients use the hosted URL with a bearer key, and clients that launch servers as local commands use the @webinarstack/mcp wrapper, which forwards to the same hosted endpoint.

Run this on your own webinar

WebinarStack records your webinar once and runs it every day after that, with follow-up driven by what each attendee actually did.