AI Agent Integration

Connect AI agents to SlideKit via MCP — generate branded slides from natural language.

AI Agent Integration

SlideKit includes an MCP (Model Context Protocol) server that gives AI agents direct access to template management, slide generation, and download.

End users can ask for slides in natural language. The agent handles the mechanical workflow behind the scenes: finding available templates, reading template analysis, selecting compatible template slides, structuring slide_data, setting rendering options, starting generation, polling status, and downloading the finished .pptx.

The user does not need to write JSON payloads by hand. The agent does need access to the MCP tools and enough template analysis context to choose the right slideId.

How agents use the API

A typical agent run looks like this:

1. List available templates.
2. Read the selected template’s analysis.
3. Choose actual analyzed template slides whose shapes match the user’s requested content.
4. Build slide_data from the user’s natural-language request and source data.
5. Call generate_slide or generate_deck.
6. Poll get_generation_status until the job is ready.
7. Return the download URL or save the .pptx with download_presentation.

Setup

The MCP server supports three transports. Pick the one that matches your environment.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "slidekit": {
      "command": "python",
      "args": ["mcp_server.py"],
      "cwd": "/path/to/powerpoint_api",
      "env": {
        "POWERPOINT_API_URL": "http://localhost:8000",
        "POWERPOINT_API_KEY": "your-api-key"
      }
    }
  }
}

Restart Claude Desktop after saving. The SlideKit tools appear in the tool picker.

Claude Code

Add a .mcp.json file to your project root:

{
  "mcpServers": {
    "slidekit": {
      "type": "stdio",
      "command": "python",
      "args": ["/path/to/powerpoint_api/mcp_server.py"],
      "env": {
        "POWERPOINT_API_URL": "http://localhost:8000",
        "POWERPOINT_API_KEY": "your-api-key"
      }
    }
  }
}

Claude Code loads this automatically when you open the project directory.

Remote agents (SSE / HTTP)

For custom agent frameworks or remote deployments, run the server in network-accessible mode:

# If the SlideKit API is also on port 8000, choose another MCP port.
export FASTMCP_PORT=8001

# SSE transport
python mcp_server.py --sse

# Streamable HTTP transport
python mcp_server.py --streamable-http

Point your agent at the server URL:

{
  "mcpServers": {
    "slidekit": {
      "url": "https://your-server.com/sse"
    }
  }
}

For streamable HTTP, use the /mcp endpoint instead:

{
  "mcpServers": {
    "slidekit": {
      "url": "https://your-server.com/mcp"
    }
  }
}

Set POWERPOINT_API_URL to the address of your running SlideKit instance and POWERPOINT_API_KEY to a valid API key. If the API is running locally on the default development port, http://localhost:8000 works.

MCP tools

The server exposes the following tools:

Tool	What it does
upload_template	Upload a .pptx template file
analyze_template	Extract slide layouts, placeholder shapes, table styles, and chart presets from an uploaded template
get_analysis_status	Poll template analysis progress
list_templates	List available templates, filtered by status or category
get_template	Get metadata for a single template
get_template_analysis	Get full analysis results — every slide’s shapes, dimensions, and placeholder positions
get_slide_details	Inspect a specific slide’s placeholders
generate_slide	Start an async single-slide generation job
generate_deck	Start an async multi-slide presentation job
get_generation_status	Poll generation progress
download_presentation	Download a generated .pptx to a local path or as base64
update_charts	Update chart data in an existing presentation without regenerating
delete_template	Remove a template

There is no separate MCP matching tool today. Agents choose slides by inspecting get_template_analysis and comparing the user’s requested content to each analyzed slide’s shapes, dimensions, tables, charts, and text regions.

Skills for Claude Code

We provide four skills that teach the agent how to use the API. Copy .claude/skills/ into your project — Claude Code loads them automatically.

• generation-workflow — template discovery, layout matching, single-slide vs multi-slide generation
• text-blocks — headers, bullets, text formatting, rendering options
• chart-blocks — bar and column charts, series, axes, legends, data rows
• table-blocks — multi-paragraph cells, logo cells, conditional formatting, pagination