McKinsey PPT Design System

Last reviewed: 2026-04-28

McKinsey Style 72 Layouts Open Source

Best for: McKinsey-style decks requiring precise layout control — strategy, executive briefings, and consulting deliverables where slide structure is as important as content

Not ideal for: Brand-matched aesthetics beyond McKinsey style, or quick decks where layout precision isn't the priority

CPU-First Architecture

80%

CPU / Deterministic

Layout selection, shape placement, palette enforcement, and QA run locally without LLM calls

20%

LLM Decisions

Content structuring, headline writing, and narrative decisions that require language understanding

This split makes outputs highly consistent — the visual system behaves deterministically, so two identical prompts produce identically laid-out decks.

72 Layout Patterns

2-col compare

3-col pillar

Full-bleed chart

Exec summary

SCR headline

Process flow

2×2 matrix

Waterfall

Timeline

Org chart

Showing 10 of 72 layout patterns. The full library covers every standard McKinsey deck type.

Core Features

📐

70 Authentic McKinsey Layouts

Every pattern in the library mirrors real McKinsey slide conventions — flat design, navy/grey palette, serif titles, action-oriented headlines. Not approximations.

🖥️

python-pptx Engine

All shapes, text boxes, and charts are native PPTX objects — fully editable in PowerPoint. No image embeds, no locked shapes.

✅

Post-Generation QA

Automatic checks after generation verify alignment, font consistency, color adherence, and text overflow — same category of QA as the Anthropic PPTX Skill.

⚡

Predictable, Consistent Output

Because 80% of the system is deterministic CPU logic, you get the same visual quality on every run. No layout drift, no "sometimes it looks great, sometimes it doesn't."

Signals

GitHub Stars

66 ⭐

Maintenance

Active

Engine

python-pptx (native)

Layout Count

72 patterns

Architecture

80% CPU / 20% LLM

QA Pipeline

Post-gen built-in

Quality Assessment

Layout Precision

4.8

McKinsey Fidelity

4.6

Output Consistency

4.7

Ease of Use

2.9

Code inspection confirmed: MckEngine wraps python-pptx with 67 high-level methods (eng.cover(), eng.timeline(), eng.donut() etc.), 72 slide types (1 retired), and a three-layer file integrity guard. QA pipeline (qa.py + run_qa_tests.py) verified. Ease of Use significantly lower than prior estimate: the previously shown python generate.py command does not exist — this is a Claude Code Skill, not a CLI tool. Dependencies heavier than displayed: rembg (pulls ~150MB onnxruntime ML runtime) and tencentcloud-sdk-python. AI cover image feature requires Tencent Hunyuan API key (primarily available to China-region users); core layout generation works without it.

Install & Use

# Install as Claude Code Skill
git clone https://github.com/likaku/Mck-ppt-design-skill
cp -r Mck-ppt-design-skill ~/.claude/skills/mck-ppt-design

# Install dependencies (note: rembg pulls ~150MB onnxruntime)
pip install python-pptx lxml Pillow numpy rembg tencentcloud-sdk-python

# Optional: set TENCENT_SECRET_ID + TENCENT_SECRET_KEY for AI cover images
# (Tencent Hunyuan API — China region; core layout works without this)

# Then use naturally in Claude Code:
# "Create a McKinsey-style strategy deck on digital transformation"
# "Build a 20-slide market entry analysis using SCR structure"
# "Generate a consulting deliverable on supply chain optimization"
# Claude writes MckEngine calls (eng.cover(), eng.timeline() etc.) and executes them
        