PPT Master tested: free AI PowerPoint slides from Claude Code

PPT Master is a free tool that, in its own words on GitHub, turns documents into "real, native PowerPoint decks". It has about 64,000 stars on GitHub. We gave it a made-up quarterly report and let it build a 6-slide deck, twice, and we let plain Claude Code do the same job without it. PPT Master's slides looked far more designed, and every number was right. But each deck took almost 5 minutes and about $1.15 of AI use, against half a minute and 7 cents without it.

What is PPT Master?

PPT Master is a project on GitHub by Hugo He, a finance professional who "regularly reviews and edits presentation decks". You don't open it like a normal program. It is a "skill": a folder of instructions and small Python programs that an AI agent reads and follows. An AI agent is an AI that can read and write files on its own, like Claude Code, Codex or Cursor. The author's own pick is Claude Code.

You give the agent a file, like a report or a PDF, and ask for a deck. The agent plans the slides, draws each one as an SVG picture made of shapes, checks them, and then turns them into a PowerPoint file. The README promises "native shapes, data-backed charts and tables", so that you can keep editing everything in PowerPoint.

It is free under the MIT licence. You only pay for the AI model the agent uses. We mentioned it before in 8 repos from Greg Isenberg's video and in GitHub Trending. This time we tested it, like we did with Matt Pocock's skills.

Our test: one report, four decks

We wrote a short report for a bike shop that does not exist: Kettingblad Bikes, third quarter of 2026. It has one table and 22 facts we could check, like revenue of €412,600, 500 bikes sold and a repair subscription of €9.95 a month.

Then we gave the same job to Claude Code four times, each time with this prompt: "Quickly generate a 6-page PowerPoint deck (.pptx) from projects/bikeshop/sources/report.md — no need to confirm with me".

• Twice with PPT Master, version 6.7.2, inside its own folder, as its README tells you to. Because the prompt says "quickly", PPT Master uses its quick mode: it skips the planning talk with you and goes straight to the slides.
• Twice without it, in an empty folder. Claude Code then picks its own way.

We used the Claude Sonnet 5.5 model at medium effort. Everything ran in an isolated Linux sandbox, not on our own computer. We did not use any image service or paid add-ons.

What the slides looked like

PPT Master's slides look like a designer made them: a headline that tells you the point, big numbers with a short note under each, colours that match, and a source line at the bottom.

The decks without PPT Master had the same facts, in the same order. But they looked plainer: four numbers in a row, a table, and slides with only a few lines of text.

Both PPT Master runs chose their own style. Run 1 went for dark green with round shapes, run 2 for off-white with navy and red. You can see more of what it makes in the author's example decks.

A real chart, or a picture of one?

For the sales by category, PPT Master's first run made a bar chart. Not a picture of a chart, but a real PowerPoint chart: the four amounts are stored inside it, so you can change them later.

The second run drew the same kind of chart, but out of separate shapes: bars, lines and labels you can move one by one, without data behind them. The README explains this: charts are shapes by default, and only become real charts with a special setting. The second run said it tried, but skipped the real chart because one of PPT Master's own checks gave a warning.

The decks without PPT Master used a plain table instead. That is correct and easy to edit, but less striking.

Is it really editable?

Yes. We opened every file with a program that lists what is inside each slide. No slide in any of the four decks was one big picture. Everything was real text, shapes, tables or charts.

• With PPT Master: 68 and 82 pieces of text, and about 60 shapes and groups of shapes per deck. That gives a nice design, but also more small parts to click through when you edit.
• Without it: 14 and 20 pieces of text and one table per deck.

None of the four decks had speaker notes, the text under a slide for the person who presents it. PPT Master can add them, the README says, but our quick prompt did not ask for them.

Did it get the numbers right?

Yes, all of them. A script looked for the 22 facts from the report in every deck: revenue, growth, the four categories, the e-bike share, the rating, the risks and the plans. It found all 22 in all four decks, and we looked through all 24 slides ourselves.

We also looked for numbers that were not in the report. The only extra ones were labels on chart axes, like "€100k", and the data stored inside the real chart, which matches the report. Nothing was made up.

Remember: our report was short and clear. A long PDF with many numbers is a harder test.

How long, and what does it cost?

PPT Master took almost 5 minutes per deck, even in quick mode. The project warns about this itself: "10–20 min for a 10-page deck", on its page about where it is not the right choice. That number is for its full mode, with planning. Without PPT Master, Claude Code was done in under a minute.

The difference comes from the work behind the scenes. With PPT Master, the AI took 35 to 43 steps instead of 5. It loaded about 17 times as much new text, such as PPT Master's own instructions, and wrote about 11 times as much. Counting everything it read, that is about 30 times as many tokens, the small pieces of text an AI reads and writes.

We ran this on an existing Claude subscription, so we did not pay these amounts. On a plan like that, it uses part of your limit instead; see our Claude Pro vs Max test.

Watch out for fonts

One surprise came when we opened the decks with LibreOffice, a free office program, on Linux. On the cover of PPT Master's first run, the title ran into the line below it, and "+14%" broke over two lines.

The facts were fine; the layout depends on the fonts. That run used Trebuchet MS and Calibri, fonts from Microsoft that our Linux machine did not have. LibreOffice used other fonts instead, and the text no longer fit. PPT Master's own drawing of the same slide, which we opened on Windows where those fonts are installed, looks clean. The second run used Arial and looked fine in LibreOffice too.

So: if you share a deck with people who use other software, like LibreOffice or Google Slides, open it there once and check that the text still fits. We could not check PowerPoint itself, because we don't have it.

The fine print

Three things we found while reading the project:

• Sponsors that resell AI. The README promotes services that sell access to Claude, GPT and other models below the normal price; one says "as low as 7% of official rates". If you use one, what you send to the AI, including your files, goes through that company and not straight to Anthropic or OpenAI. We did not use or check these services.
• One part is not MIT. To read PDF files, PPT Master uses PyMuPDF, which has a stricter licence called AGPL-3.0. It is only needed for PDF input. The project's own notes ask you to check those terms if you pass PPT Master on to others together with that part.
• Same name, other sites. When we searched Google for "ppt master" on 19 October, two other sites showed up next to the project: pptmaster.app and pptmaster.online. Neither links to the GitHub project. The first is a separate AI slide maker with free and paid plans; the second is a list of AI slide tools, whose questions page mentions the GitHub project without linking to it. If you want this tool, go to the GitHub page itself.

Should you use it?

If you want a deck that looks designed, from a document you already have, PPT Master is impressive. It got every fact right and made slides you can keep editing. It is free, and it works with an AI agent you may already use.

But it is not quick, and the AI use adds up: about $1.15 for six slides in our test. If you only need the content on slides and will style them yourself, asking Claude Code directly gave the same facts in half a minute for about 7 cents.

Whichever you pick, open the file before you present it. None of our four runs looked at its own slides; they all said so in their summary. And ask for speaker notes if you want them.

How we checked

On 19 October 2026 we downloaded PPT Master version 6.7.2 into a Vercel Sandbox, a separate Linux computer in the cloud, and installed what it needs. We ran Claude Code 2.1.303 there four times with Claude Sonnet 5.5 at medium effort, on the site owner's Claude subscription: twice inside the PPT Master folder and twice in an empty folder, with the same prompt and the same made-up report. Time, steps, tokens and the model-use amounts come from Claude Code's own report of each run; the amounts are what the run would cost at list prices.

We opened every deck with python-pptx to count text boxes, shapes, charts, tables, pictures and notes, and checked the 22 facts on every slide. We turned the slides into images with LibreOffice in the same sandbox, and opened PPT Master's own SVG drawings of the slides on Windows, where the right fonts are installed. The slides in this article are those images. The decks without PPT Master were built with python-pptx; their file details still carry its default template.

We also read PPT Master's README, its page on where it is not the right choice and its notes on PDF reading, and visited pptmaster.app and pptmaster.online. The charts and diagrams were made by us, and none of the images was made by AI. This was a small test with one short report. PPT Master changes almost every week, so your results may differ.