MCP connector — connect your AI directly to Builder

A Business Premium feature (or for a partner whose admin has enabled the MCP connector feature). MCP (Model Context Protocol) is an open standard that lets AI clients — Claude Code, Claude Desktop and others — connect to external services. With the MCP connector your AI works directly in Builder: it creates presentations, designs templates and translates finished presentations — it fetches the guide by itself, performs the task and returns a link to the editor. No manual JSON copying.

In the MCP dialog (button on the presentations list) pick at the top which client you use. The recommended path is Claude Desktop / web — you connect by signing in, with no token and no terminal.

Claude Desktop / web (easiest — via sign-in)

What you need: the Claude Desktop app or claude.ai web (Pro/Max plan). No token, no terminal, no install.

Open connectors — in Claude: Settings → Connectors → Add custom connector.
Paste the Builder address:
```
https://pmgclub.com/api/mcp
```
(You'll also find the address in the MCP dialog in Builder, with a copy button.)
Sign in and approve — click Connect → sign in to PMG Club → on the consent screen click Approve and continue.
Done — Claude sees the Builder tools. You can revoke access anytime in Claude (the Disconnect button) or under Connected apps in your account.

Gemini (basic set — via sign-in)

Gemini connects the same way as Claude (by signing in), but to a different address — the basic set address. Gemini cannot handle Builder's full tool catalog (it validates tool descriptions more strictly and limits their total size), so it only gets what is needed to create content.

Open connected apps — in Gemini (gemini.google.com): Settings → Connected apps → Custom apps → Add. ⚠️ Availability (as of October 2026): Google offers custom MCP apps only in the US, only for a personal Google account (not work/school), with "Keep Activity" on and an AI Pro / AI Ultra subscription. Work accounts go through Gemini Enterprise (admin → Connected apps → Add MCP server). If you don't see the option, it is a Google limitation, not your PMG Club account — our connector is ready and will work as soon as Google makes the feature available.
Paste the basic-set address:
```
https://pmgclub.com/api/mcp/zakladni
```
(Also in the MCP dialog in Builder under the Gemini option.)
Sign in and approve — click Next → sign in to PMG Club → Approve and continue.
Done — Gemini sees the Builder tools. Revoke access in Gemini (Disconnect) or in Connected apps on your account.

What the basic set contains: creating and editing presentations (guide, designs, new presentation, adding/reworking slides, presentation description, folders), translation, X-elements (builder_x_guide, builder_x_add, builder_x_update, builder_x_describe, builder_animate_slide), image and video placeholder, video podcast (guide, hosts, script, line edits, status — voices and voicing then in the app; the Video podcast feature must be enabled) and help. What it does not contain: music (Studio3), drawn podcast characters, Q-Live, template design (import_design) and M-icon — use Claude or ChatGPT with the full address https://pmgclub.com/api/mcp for those.

You can use the basic-set address in any other AI client that struggles with a large catalog — or in Claude when you want to offer the AI content creation only.

Claude Code (in the terminal — via token)

For the terminal AI assistant. What you need: a Claude account (Pro/Max) and the Claude Code app.

Open a terminal — Mac: Cmd+Space → type "Terminal" → Enter; Windows: PowerShell
Install Claude Code (once) — paste and confirm with Enter:
```
npm install -g @anthropic-ai/claude-code
```
If the terminal doesn't know npm, install Node.js from nodejs.org (LTS button) first and reopen the terminal.
Sign in — type claude, Enter, finish the sign-in in the browser that opens, then exit Claude with /exit
Create a token — in Builder click MCP → the Claude Code tab → Create token. The token is shown only once; the dialog offers a ready-made command:
```
claude mcp add --transport http builder https://pmgclub.com/api/mcp --header "Authorization: Bearer YOUR_TOKEN"
```
Connect Builder — paste the copied command into the terminal and confirm with Enter
Verify — run claude and type /mcp; the list should show builder — connected

Example prompts

Write naturally; Claude handles everything itself and returns a link to the editor:

"Create a Builder presentation about our loyalty program for cafés, 8 slides, with voice-over texts."
"Design a Builder template in my brand colours (dark blue #0E1B2C and gold), slug my-brand. Then create a presentation in it about the summer offer."
"Create a Builder presentation in the my-brand design about the new price list."
"Translate my presentation 'Summer offer' into German." (slide texts and voice-over scripts; saved as a second language in the same file)
"Create a new presentation about Q4 results in the same style as my 'Company template' presentation." (reuses the look of an existing presentation)
"List my Builder presentations."

What the AI can do through the connector

Tool	Description
`get_story_guide`	fetches the current guide and format rules
`list_designs`	lists designs you are entitled to (incl. restricted ones)
`import_story`	creates a new editable presentation (optionally into a folder), returns a link
`add_slides_to_presentation`	adds slides to an existing presentation (at the end or after an index)
`replace_slides`	replaces a range of slides — destructive, with a backup
`list_presentations`	lists your presentations (incl. which folder)
`list_folders`	lists your Builder folders
`create_folder`	creates a folder
`get_design_guide`	fetches the guide for designing a look (template)
`import_design`	creates a template draft — you polish it in the editor
`get_translation_guide`	fetches the guide for translating a presentation
`get_presentation_texts`	pulls a presentation's texts to translate (slide texts, voice-over scripts, video subtitles)
`set_presentation_translations`	saves the translation as a second language in the same file (overlay)
`create_translated_copy`	creates a standalone translated copy — any language (incl. Vietnamese)
`get_qlive_guide`	fetches the guide for creating a Q-Live (live poll/quiz)
`create_qlive_event`	creates a live poll/quiz (draft) with questions, returns a join code
`list_qlive_events`	lists your Q-Live events
`add_qlive_to_presentation`	embeds Q-Live (QR / live results) straight into a presentation — new or existing slide
`get_studio_guide`	fetches the guide for creating a song in Studio3 BOX
`compose_song`	creates a song in Studio3 BOX from a musical intent (style/mood/blocks)
`list_songs`	lists your Studio3 BOX songs
`set_song_expression`	sets a song's expression — velocity and aftertouch per voice plus section dynamics (audible in the Studio, written to the MIDI export)
`help_list`	lists the contents of this help (groups and pages)
`help_search`	searches the help — returns pages with a link and a snippet
`help_get_page`	fetches a whole help page — the AI advises you exactly by it

Deliberately nothing more: the AI has no access to deleting, exports or sharing.

AI as a designer: via import_design your AI can also propose a whole look — it arrives as a template that you polish in the editor (the final word is yours), and with a slug the AI uses it as the design for content from then on. Example: "Design a Builder template in my brand colours (dark blue + gold), then create a presentation in it about…"

Translating a presentation with your AI

Your AI can also translate a finished presentation — for free, because it does the translation itself. The translation is saved as a second language in the SAME file (the Czech original stays), exactly as if you clicked Translate in the editor. It covers slide texts, voice-over scripts and video subtitles.

Just say, e.g.: "Translate my presentation 'Summer offer' into German." The AI fetches the guide, pulls the texts, translates them and writes them back. Then:

Open the presentation in the editor.
Switch the language to the target (e.g. DE) — you'll see the translated version.
Review and tweak by hand if needed (your corrections take priority).
For voice-over, generate the audio for the translated script in the editor.

Supported languages (overlay with a switcher): English, German, French, Spanish, Italian, Polish, Slovak, Ukrainian.

Or a standalone copy (any language). For a full localized version for a market — e.g. Vietnamese — say: "Create a Vietnamese copy of my presentation." The AI clones the deck and replaces the texts with the translation → a new standalone presentation (the original stays intact) that you edit, dub and export independently. Works for any language. The copy is a snapshot — later edits to the original don't update it.

Q-Live (live poll/quiz) via AI

Through the connector your AI can also create a Q-Live — a live poll or quiz the audience joins with a code and votes on from their phones. Say e.g.: "Create a Q-Live quiz for the payments slide, 5 questions." The AI creates the event as a draft (nothing runs until you start it live) and returns a 6-digit join code.

Question types: multiple choice, quiz with a correct answer, rating scale, word cloud, or open text.

Embedding — automatically: the AI can embed the Q-Live into the deck itself (add_qlive_to_presentation) — either as a new standalone slide (heading + QR/results) or onto a specific existing slide. So "Build a presentation about payments and add a Q-Live quiz at the end" works in one prompt. You can reposition the widget later in the editor.

Or manually: in Builder add a Q-Live (live results) or Q-Live QR object and pick this event by the returned code.

The slide shows the join QR/code and live results. The event activates only when you run it live — until then it consumes nothing.

Especially handy when you build the whole presentation via your own AI: let it prepare the polls from the content context and embed them right away.

Editing an existing presentation and backups

Through the connector the AI can also edit a finished presentation — add slides (add_slides_to_presentation) or replace a range of slides (replace_slides). So the AI can never overwrite your work irreversibly:

A backup of the previous state is saved automatically before every MCP edit (this also covers translations and Q-Live embedding).
The editor shows a yellow bar (currently in Czech: „⚠️ Tato prezentace byla upravena přes MCP konektor…") with a Zálohy (n) (backups) button — you see a list with time and reason and restore any version with Restore. (A restore itself is backed up too, so nothing can go wrong.)
Replacing (replace_slides) is destructive — the replaced slides lose their manual design. The AI is instructed to ask you first and warn that the presentation may change significantly.

The last ~20 backups per presentation are kept.

Studio3 BOX — music via AI

Through the connector your AI can also create a song in Studio3 BOX. It doesn't compose note by note — it just provides the intent (style, mood, length) and the built-in engine composes the music. Say e.g.: "Create an upbeat dance track in Studio3, 8 blocks."

Styles: vzlet (melodic/building), party (upbeat dance), miniatura (shorter, calmer). Mood light/dark, length 4/8/12 blocks.
Returns a DNA (short code) — the same DNA = the same song, so you can save it or make variations.
Beyond the intent you can ask the AI to tune the sound: drum kit, voices (pad/lead/bass), BPM, delay and especially reverbs — master reverb plus a separate reverb per voice (drums/chords/melody/bass), either as a simple amount or a full reverb (length, predelay, filters, reverse). E.g. "…with a large reverb on the melody and a piano lead."
You can also fine-tune each instrument's synth parameters (the "🎛 Sound" sliders) — e.g. "e-piano less metallic with a longer ping, strings as a big section." The AI fetches the exact parameters and ranges from the guide.
The song appears in Studio3 BOX → My songs; open it there to play, tweak further and export (WAV is Premium).

Credits and costs

Presentation content and translations are created by your AI in your account (Claude, ChatGPT…) — PMG Club credits are not consumed when working via MCP. The Translate button inside the editor uses our AI and costs credits as usual; a translation via your own AI (MCP) is free. Generating voice-over audio in the editor is always credit-based.

Token security

A token is valid until you revoke it (trash button in the MCP dialog)
Only a fingerprint is stored in the database — once the dialog closes, nobody can display it again
Each token shows when it was last used
Never share the token — whoever has it can create presentations on your behalf

Alternative without MCP

If your AI does not support MCP, Import from AI with the downloadable guide still works — same format, you just transfer the file manually. See the "Import from AI (Story JSON)" help page.