SlideSpeak

Converts natural language descriptions into structured presentation definitions through an MCP (Model Context Protocol) server that translates user intent into slide schemas. The system parses free-form text input describing presentation content, structure, and styling, then generates PowerPoint-compatible output by mapping semantic intent to presentation primitives (slides, layouts, text blocks, formatting). Uses MCP's standardized tool interface to expose presentation generation as a callable resource that LLM agents can invoke with structured parameters.

Maps natural language descriptions of slide content to predefined PowerPoint slide layouts and content placeholders through semantic understanding of intent. The system infers appropriate layout types (title slide, bullet list, two-column, image+text, etc.) from content description, then populates placeholders with generated or provided text. Uses python-pptx's shape and text frame APIs to position content within layout constraints, handling text wrapping, font sizing, and placeholder alignment automatically based on layout schema.

Manages the creation of multi-slide presentations by decomposing a high-level presentation goal into individual slide definitions, sequencing them logically, and orchestrating their generation into a cohesive PowerPoint document. The system handles slide ordering, cross-slide consistency (fonts, colors, branding), and logical flow between slides. Uses python-pptx's Presentation object model to manage slide collections, apply master slide formatting, and ensure consistent styling across the entire deck through centralized theme and layout management.

Generates presentation content (titles, bullet points, speaker notes, descriptions) by prompting an LLM with structured instructions that specify content requirements, tone, length, and format. The system constructs detailed prompts that guide the LLM to produce content suitable for specific slide types and audience contexts. Uses prompt engineering patterns to ensure consistent output format (e.g., bullet lists with 3-5 items, title length under 10 words) and semantic coherence across generated content. Integrates with MCP's tool interface to expose content generation as a callable capability that agents can invoke with parameterized prompts.

Generates valid PowerPoint (.pptx) files that are compatible with Microsoft Office, Google Slides, and other presentation software by using python-pptx library to construct Office Open XML (OOXML) structures. The system builds presentation objects with proper XML serialization, handles embedded resources (fonts, images, color schemes), and ensures compliance with PowerPoint format specifications. Manages file I/O, temporary file handling, and output path configuration to reliably produce downloadable or storable presentation files.

Exposes presentation generation capabilities as MCP tools with standardized schema definitions that enable AI agents to discover, understand, and invoke presentation generation functions. The system defines tool schemas specifying input parameters (presentation topic, slide count, audience, style), output format, and tool descriptions in MCP's standardized format. Agents can parse these schemas to understand what parameters are required, what types are expected, and what the tool produces, enabling autonomous tool selection and invocation without hardcoded integrations.

Applies consistent visual styling and branding to presentations through PowerPoint master slide templates that define color schemes, fonts, logos, and layout standards. The system loads predefined master slides from template files, applies them to generated presentations, and ensures all slides inherit the master formatting. Uses python-pptx's theme and layout APIs to apply master slide styling, manage color palettes, and enforce font consistency across the entire presentation without requiring per-slide styling configuration.