MCP for Presentations in Claude and Cursor

Your agent already writes the deck. The slow part is everything after: save the files, open a browser, upload, copy a URL, paste it back. Model Context Protocol (MCP) removes that loop. Claude or Cursor create the deck, publish a new version, and hand you a share link as you, without you leaving the editor.

Here is what you get:

• One link that updates in place. Say “update the numbers” next week and the client’s URL shows the new version.
• Publishing as a tool call. No export, no ZIP, no “which v3.pptx did I send.”
• The same deck you already have. Prezzly hosts and versions your HTML. It does not generate slides, and it does not call a model of its own.

If you already send an AI-made HTML deck by drag-and-drop, MCP is that same job with the download step deleted.

What you actually get (in plain English)

Without MCP, Monday looks like this: generate files, save them, open a browser, upload, copy a URL, paste it back into the chat. Every revision repeats the tour.

With MCP, you stay in the editor. You say “tighten the chart on slide 4, publish a new version, give me the view URL.” The agent does the filing. You send one link to the client. Next week you say “update the numbers” and the same link shows the new version. That is the outcome for a marketer or consultant: no ZIP, no second tool, no lost versions.

The agent acts as you, not as a magic write button inside a slide editor. It can also read and rewrite speaker notes, review the questions your audience left after a call, and build in your own saved styles. You do not need to learn any commands. You need a connection once, then a clear prompt.

For a consultant shipping client work, that is the difference between a delivery ritual and a sentence. You finish the deck in the editor, ask for a link, and drop it in the thread. When feedback lands, you fix it in the same chat and the client’s link is already current. Nothing leaves the workflow you are already in, and there is no separate “publishing” chore at the end of the day.

How to connect

Open Connections while signed in. You set this up once per tool.

Claude in the browser. Add Prezzly as a custom connector, sign in, and approve access. From then on, any chat can publish for you.

Cursor, Claude Desktop, and similar tools. Generate a key on the Connections page and paste it into your client’s settings. Treat the key like a password, and revoke it if a machine is shared.

That is the whole setup. After it, publishing is something you ask for in plain language.

What you can ask the agent to do

You do not have to learn a command list. In plain terms, you can ask the agent to:

• Publish a folder as a presentation or a dashboard, and give you the share link.
• Update an existing deck in place so the same link shows the new version.
• Create or revoke a time-limited view link.
• Grant a specific person read access, so you can share a Claude artifact privately instead of posting a public link.
• Read, rewrite, or clean up your speaker notes.
• Pull the questions your audience left, so you can decide what to change.

One thing worth saying in the prompt: whether you want a presentation (slides) or a dashboard (a scrollable report). If you do not say, you get slides. For a shareable live dashboard, ask for a dashboard explicitly.

How to prompt so you get clean results

Ask the agent to update the deck you have, not to start over. That keeps one link and one version history instead of a pile of near-duplicates.

Good prompts:

• “List my presentations. If Q3 review exists, update it. If not, create it.”
• “Open Q3 review, make the hero title shorter, publish the new version.”
• “Create a 7-day view link and paste it here.”

Bad prompts:

• “Make me a new Prezzly called Q3 review v2” every time you change a sentence. You will drown in duplicate links.
• “Rebuild this from scratch in Slides.” You already have the deck - host and version the HTML you generated instead.

If you host Reveal.js slides, the same rule applies: ask the agent to publish the whole built deck, then update it in place.

How MCP fits your workflow

• Your agent writes the deck; Prezzly hosts, versions, and shares it. That division of labour is the point: you keep the tool you like for creating, and get a durable link for delivering.
• You stay in control of every revision. The agent can pull the questions your audience left after a call, and you decide what to change - the deck only updates when you ask.
• It works alongside any meeting tool. Paste the share link into Meet, Teams, Zoom, or Slack and it opens like any URL.

Frequently Asked Questions (FAQ)

Do I need MCP to use Prezzly?

No. You can upload from the landing page or your library. MCP is for people who already work in Claude or Cursor and want publishing to be a prompt instead of a chore.

Is this official Claude support?

You add Prezzly on the Connections page as a custom connector or with a key. That is the supported path in the product.

Can ChatGPT use the same connection?

If your client speaks MCP and can sign in or use a key, it can publish the same way. The setup instructions call out Claude and Cursor, but the approach is not exclusive to them.

Will the agent overwrite my deck by mistake?

Updates create a new version and keep the old ones, so you can restore an earlier version. Ask the agent to update the existing deck rather than create duplicates.

Does connecting an agent train a model on my decks?

Prezzly does not run its own model on your files. Your agent is whatever tool you already use, so read that vendor’s terms. Prezzly does not train on your uploads.

---

Ready to skip the ZIP dance? Open Connections, add Claude or Cursor, and let the next version publish as a prompt.