Auto-Generate Slides from Your Meetings with SuperIntern MCP and the Claude Code pptx Skill

Some meetings end when you close the call window. Client meetings don't: they end an hour later, after you've written the internal report deck.

You know the drill. The conversation is still fresh, but the slides won't build themselves. What was their actual pain point? What did we promise? You reconstruct it from notes into a PowerPoint file, slide by slide, and the polish you can afford depends on how much else is on fire that day. Interviews are the same: the debrief document always takes longer than it should.

Here's the thing: if you already use Claude Code, you have a slide factory sitting in your terminal. This guide shows how to combine SuperIntern MCP (your meeting data) with the official pptx skill (PowerPoint generation), so that turning a meeting into a report deck is one prompt, and the output is a real .pptx file in your folder.

⚠️ This article was independently compiled based on publicly available information and user feedback as of August 2026.

Table of contents

• What we're building: before / after
• How it works: three roles
• Setup
• Demo 1: An internal report deck right after a client meeting
• Demo 2: An interview scorecard deck with evidence from the transcript
• Demo 3: One weekly deck from all your meetings
• Tips and caveats
• Going further: the same pattern, other tools
• FAQ

What we're building: before / after

Before: After each client meeting or interview, you rebuild the conversation from memory into slides. It costs 30 to 60 minutes, and it competes with everything else on your plate, so it slips, and the deck your team gets is thinner than the meeting deserved.

After: The meeting ends. In your terminal, you ask Claude Code once: "make the report deck from that meeting." Claude Code pulls the AI summary through SuperIntern MCP, decides what belongs on each slide, and the pptx skill writes a structured .pptx file to your folder. You open it, adjust a few words, and share it.

The raw material is always there because SuperIntern is a botless desktop meeting assistant: no bot joins your call, and transcripts and AI summaries accumulate automatically on any meeting platform, even in person. And because the output is a plain PowerPoint file on your machine, it opens in PowerPoint, Keynote, or Google Slides, and nothing leaves your computer unless you send it.

How it works: three roles

MCP (Model Context Protocol) is an open standard for safely connecting AI assistants to external tools; SuperIntern's MCP exposes your meeting data to AI clients read-only. Agent Skills are the other half: folders of instructions and scripts that teach Claude a repeatable task. The official skills repository ships the document skills that power Claude's document capabilities, including the pptx skill for creating and editing PowerPoint files.

In other words, SuperIntern provides the record, Claude Code makes the judgment calls, and the pptx skill does the manual labor. Your SuperIntern data can never be modified by the AI (the MCP has no write operations), and the only thing created is a file in your working directory.

Setup

1. Connect SuperIntern MCP to Claude Code

In your terminal, run:

claude mcp add --transport http superintern https://mcp.app.super-intern.com/mcp

Then inside Claude Code, run /mcp, select superintern, and authenticate. A browser window opens; sign in with your SuperIntern account and grant access. No API keys involved. SuperIntern-side requirements (Plus plan or higher for personal workspaces, admin enablement on Enterprise) are covered in our MCP setup guide.

2. Add the pptx skill

Anthropic publishes its document skills in the official anthropics/skills repository. Inside Claude Code, add it as a plugin marketplace and install the document skills:

/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills

That's it. Skills load automatically when they're relevant: mention "make a pptx" in your prompt and Claude Code picks up the skill on its own. You can check what's installed anytime with /skills.

To test both pieces, try:

List my 3 most recent SuperIntern meetings

If titles and dates come back, you're ready.

Demo 1: An internal report deck right after a client meeting

The classic case: a client or prospect meeting ends, and your team needs the internal readout. Fix the structure in the prompt, and every readout you produce has the same shape.

Get the AI summary of the meeting "Acme Corp intro meeting" that just ended,
and create an internal report deck as acme-readout.pptx with 5 slides:
1) Title: meeting name, attendees, today's date
2) Their situation: context and what prompted the meeting
3) Their pain points, quoting their actual words where possible
4) What we proposed and how they reacted
5) Next actions with owners and dates
Keep each slide to 4 bullets or fewer. Mark anything ambiguous
as "to confirm" instead of presenting it as settled.

Claude Code fetches the summary, drafts the content, and the pptx skill writes the file:

Created acme-readout.pptx (5 slides).

• Their situation: meeting notes currently depend on one person; evaluating tools ahead of the EU team expansion
• Pain points: "the notes bottleneck is one person", "security review is strict about recording data"
• Our proposal: 2-week trial with the CS team; SSO confirmed as a rollout precondition
• Next actions: security questionnaire by Aug 28 (us), trial start decision by Sep 4 (their side)

Anything marked "to confirm" is on slide 4. Open the file to adjust wording before sharing.

Thirty seconds after the call, the readout exists as a file. The difference from a memory-built deck is that the pain-point slide quotes what the client actually said, because it was built from the record.

Demo 2: An interview scorecard deck with evidence from the transcript

Hiring debriefs are the meeting type where memory is least trustworthy and evidence matters most. After an interview, generate a scorecard deck where every rating points at an actual quote, by sending Claude Code to the transcript, not just the summary.

Get the transcript of today's interview "Backend engineer - candidate Alex".
Create scorecard-alex.pptx for the hiring debrief:
1) One slide per evaluation axis: technical depth, system design,
   communication, ownership
2) On each slide: a 1-5 rating suggestion, 2-3 supporting quotes
   from the transcript, and any red or yellow flags
3) Final slide: points the next interviewer should probe
Do not invent quotes. If an axis wasn't covered, say "not assessed"
instead of guessing.

The "not assessed" instruction matters: the honest gap is exactly what the next interviewer needs to see. Because the deck is built from the transcript, the debrief discussion shifts from "I felt they were strong" to "here's what they actually said", and the panel argues about evidence instead of impressions.

Demo 3: One weekly deck from all your meetings

Working across multiple meetings is where MCP shines. On Friday, collect the week into a single reporting deck.

List my meetings from this week in the "Sales" project and get each
AI summary. Create weekly-report.pptx:
1) One overview slide: pipeline highlights and risks this week
2) One slide per account: status, what moved, next step
3) Final slide: where I need help from the team
Same format as last week's file if it exists in this folder.

That last line is very Claude Code: it can look at last week's file sitting in the same folder and keep the format consistent. Run the same prompt every Friday and your weekly reporting becomes a reviewed-and-sent routine instead of an evening chore. If you find yourself reusing the prompt, save it as your own skill (a SKILL.md in ~/.claude/skills/) and it becomes a one-word command.

Tips and caveats

Fix the slide structure in the prompt. A stable skeleton ("situation / pain / proposal / next actions") keeps every deck consistent and makes the extraction more reliable. Once it works, promote the prompt into a personal skill so the whole workflow is /meeting-report.

Choose summary or transcript deliberately. The AI summary is right for readouts and weekly reports; the transcript is right when evidence matters, like interview scorecards or contested decisions. Asking for both costs nothing but tokens.

Treat the deck as a first draft. The pptx skill produces a clean, structured file, but the five minutes you spend on wording and emphasis is what makes it yours. Starting from 80% is the win.

The file stays local until you share it. The deck is written to your working directory. Nothing is uploaded anywhere, and no external design service is involved; for sensitive meetings like interviews, that's the property you want.

Improve accuracy on the SuperIntern side. Names, product terms, and technical vocabulary transcribe better when they're in SuperIntern's custom dictionary, and the deck content inherits that accuracy.

Going further: the same pattern, other tools

• Prefer a cloud design tool? The same workflow with Canva as the action target, generating an editable deck in your Canva account, is covered in the Canva article. Rule of thumb: pptx skill for local files and full control, Canva for visual polish and sharing
• Claude Code for implementation: turning spec meetings directly into code is covered in the spec-to-code article
• Follow-up emails: drafting the email that carries your deck, from the same meeting data, is covered in the Gmail article

FAQ

Q. Can I use SuperIntern MCP on the free plan?
A. No. Personal workspaces require the Plus plan or higher. Team plan workspaces can use it as is, and on Enterprise it works once an admin turns on "MCP access" in workspace settings.

Q. Do I need PowerPoint installed?
A. No. The pptx skill generates a standard .pptx file, which opens in PowerPoint, Keynote, Google Slides, or any compatible tool.

Q. Can the AI modify my meeting notes or transcripts?
A. No. Every SuperIntern MCP tool is read-only; no create, edit, or delete operations exist. The only thing written is the deck file in your local folder.

Q. Where does my meeting data go?
A. Claude Code reads it through SuperIntern MCP under your own login, and the generated deck is a local file. No design SaaS or third-party storage is involved in this workflow.

Q. Does this work in AI clients other than Claude Code?
A. SuperIntern MCP itself works with MCP-capable clients like Claude, Codex, and Cursor. The pptx skill shown here is an Agent Skill from the anthropics/skills repository; on other clients, check their current support for skills or generate documents with the tools they provide.

Q. What's the difference from the Canva integration?
A. The pipeline is identical (SuperIntern MCP provides, the AI judges); only the action target differs. The pptx skill writes a local file with no external service, while Canva's MCP creates a design in your Canva account for visual editing. Pick per deck.

SuperIntern already keeps the record of every meeting. Your terminal can now turn that record into the deck your team is waiting for.

Try SuperIntern Free