How to Turn Markdown and Code into Interactive Presentations with AI Agents
A practical guide to turning Markdown specs, RFCs and project notes into interactive web decks or MP4 explainers in Claude Code, Codex and Cursor — without losing numeric accuracy
Technical teams write everything in Markdown: architecture RFCs, release notes, post-mortems, and quarterly metrics. Turning those notes into a presentation usually means copying paragraphs into a slide tool, redrawing charts by hand, and watching numbers drift out of sync with the original file.
An agent-native workflow flips that process. Because your coding agent already has local file access, it can parse the source document directly, verify every figure against the original text, and assemble a playable HTML deck or timeline composition on your own machine.
Why static slide exports fall short for technical content
Most AI presentation tools generate static bullet cards and generic stock illustrations. When you present an engineering RFC or a revenue breakdown, three problems show up immediately:
- Rounded or invented metrics. Generative slide writers frequently paraphrase
$14.82M (-3.4%)into "roughly $15M," breaking trust in review meetings. - Dead charts and screenshots. Viewers cannot filter a series, inspect a tooltip, or verify which section of the source report produced a number.
- One-shot layouts. Without a real timeline or structured speaker script, adjusting pacing or exporting a clean 16:9 or 9:16 MP4 requires starting over in a video editor.
The four-step Markdown-to-Motion pipeline
With the claudemotion CLI and skill installed in Claude Code, OpenAI Codex, or Cursor, the entire pipeline runs locally:
npx -y claudemotion@latest init ./claudemotion/q3-review
npx -y claudemotion@latest ingest ./docs/q3-metrics.md --project ./claudemotion/q3-review --lang en
npx -y claudemotion@latest pick-template --project ./claudemotion/q3-review
npx -y claudemotion@latest compose slides.json --project ./claudemotion/q3-review
npx -y claudemotion@latest check --project ./claudemotion/q3-review
1. Structured ingestion (content.json)
claudemotion ingest extracts every heading, point, table, number, sign, and source line reference into content.json. Nothing is summarized away before you choose a format.
2. Choosing PPT or Motion
- PPT mode covers every chapter of your Markdown file, writes a spoken-style speaker script (
narration.md) for every slide, and mounts an interactive slide player with optional ElevenLabs voiceover. - Motion mode maps scenes onto a HyperFrames timeline with spring easing, 2.5D camera framing, animated ECharts, interactive quizzes, and deterministic MP4 export.
3. Adding interactive verification blocks
Instead of pasting static screenshots, slides can include live runtime components:
- Source citations (
.cm-cite): viewers click any KPI to read the exact quote and section from your original Markdown file. - What-if sliders (
.cm-slider): stakeholders drag a parameter slider to test sensitivity assumptions live in the browser. - Knowledge checks (
.cm-quiz): educators and developer advocates embed single or multi-choice questions with instant explanations.
4. Deterministic verification (claudemotion check)
Before opening the local studio editor or rendering MP4, claudemotion check verifies slot capacities (so text never overflows or shrinks arbitrarily), checks chart schemas, and confirms every number on screen exists verbatim in content.json.
Get started in your terminal
Browse rendered outputs on the style examples page, generate a structured prompt with the free prompt generator, or install the skill in the AI coding agent you already use.
Note: Claude Motion (claudemotion.design) is an independently operated product and has no relationship with Anthropic, PBC or Claude.