slide-deck-mcp
Enables creating and managing AI-powered slide decks on deck.4hum.ai with tools to save, update, generate images, and list decks.
README
slide-deck-skill
An agent skill package for creating, editing, and publishing professional slide decks on deck.4hum.ai.
The agent generates slide JSON directly from the slide-scene-graph v0.4.0 schema, validates it locally, then saves it through the deck.4hum.ai REST API via bundled Python scripts. No backend LLM call is needed for content — the agent writes the deck JSON itself.
What it does
- Designs a custom theme matched to the topic's tone and audience (not just picks from a list — uses two structural reference themes as a starting point)
- Generates charts, tables, diagrams (Mermaid), images, and rich text objects
- Validates the deck JSON locally before any network call
- Previews rendered slides as screenshots so the agent can evaluate and iterate
- Saves to deck.4hum.ai and returns the edit URL
- Updates existing decks via
update_deck.py - Generates custom hosted images via the API
Prerequisites
- Python 3.8+
- Network access to the deck.4hum.ai API
- A deck.4hum.ai account (run
python scripts/auth.pyonce to authenticate)
Installation
The easiest way — open a session with your agent and ask it to install the skill globally:
"Clone https://github.com/4hum-ai/slide-deck-skill and install it globally so the skill is available in all future sessions."
The agent knows where its own global skills directory is and will handle the rest. Start a new session afterward to activate it.
Manual install (Claude Code)
If you prefer to do it yourself:
# Global — available in all Claude Code sessions:
git clone https://github.com/4hum-ai/slide-deck-skill ~/.claude/skills/slide-deck-skill
# Project-local — available only in this project:
git clone https://github.com/4hum-ai/slide-deck-skill .claude/skills/slide-deck-skill
Start a new session — Claude Code auto-discovers skills from those locations.
To update: cd ~/.claude/skills/slide-deck-skill && git pull
Updating the skill
Update the skill itself (get the latest version)
# If installed via git clone:
cd ~/.claude/skills/slide-deck-skill # or wherever you installed it
git pull origin main
Check CHANGELOG.md for what changed between versions.
If a new version has schema or script changes, re-authenticate and re-validate any decks you intend to update:
python scripts/auth.py
python scripts/deck_validator.py < my-deck.json
Update an existing deck's content
To push new JSON to a deck you already created:
# By deck ID (printed when you first saved the deck):
python scripts/update_deck.py "<deck-id>" < updated-deck.json
# Or pipe from a generator script:
python my_generator.py | python scripts/update_deck.py "<deck-id>"
The update script validates the JSON before sending, same as save. The deck
URL does not change — reload https://deck.4hum.ai/app/decks/<id>/edit to
see the new content.
Bump the skill version (maintainers)
When you make changes to the skill itself (scripts, references, SKILL.md),
update metadata.version in SKILL.md and add an entry to CHANGELOG.md
before tagging a release.
Quick start
Once installed, just ask your agent:
"Create a slide deck on the benefits of microservices architecture."
The agent will design a custom theme, plan the slides, generate deck JSON, validate it, preview the result, and return a link.
Structure
slide-deck-skill/
├── SKILL.md # Agent instructions (loaded at activation)
├── scripts/
│ ├── auth.py # Credential management
│ ├── block_builder.py # Block builders: card, portrait_card, kpi_card, grid, chart, table…
│ ├── deck_patterns.py # Backwards-compat re-export shim (deck_patterns → block_builder)
│ ├── deck_validator.py # Local preflight validator (--strict quality pass)
│ ├── preview_deck.py # Structural summary + WCAG contrast check (--theme-check)
│ ├── patch_slide.py # Targeted single-slide edit / insert / delete
│ ├── generate_image.py # Image generation
│ ├── save_deck.py # Create deck via API
│ └── update_deck.py # Update existing deck
├── references/
│ ├── scene-graph.md # Full deckJson schema reference
│ ├── theme-presets.md # Theme JSON objects, color tokens, font mood table
│ ├── commands.md # Script commands, env vars, image sizing guide
│ └── image-prompts.md # Proven prompt templates (portrait, hero, icon, dataviz)
└── examples/
└── agent_skills_marketplace.py # Complete five-slide example
License
MIT — see LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。