Docs, MCP, Multimodal RAG are Live

What's New

Three major platform additions ship today: comprehensive API documentation, Model Context Protocol (MCP) integration, and multimodal RAG support for images, PDFs, Word documents, and PowerPoint files.

New Documentation

• Interactive API playground — test every endpoint directly in the browser with your API key.

• Quickstart guides for Python, TypeScript, and cURL with copy-paste examples.

• Architecture diagrams showing how memory, RAG, and routing interact internally.

• Migration guides from OpenAI Assistants, LangChain, and LlamaIndex.

MCP Integration

• Backboard now exposes a Model Context Protocol server for tool-use workflows.

• Connect any MCP-compatible client (Claude Desktop, Cursor, custom agents) to Backboard.

• Expose memory search, document retrieval, and thread management as MCP tools.

• Agents can read and write memories, query RAG, and manage state through MCP.

Multimodal RAG

RAG now supports non-text content natively:

• Images: Extract text via OCR, describe visual content, embed for semantic search.

• PDFs: Full text extraction with layout preservation, table parsing, and figure detection.

• Word documents: .docx parsing with heading structure, styles, and embedded images.

• PowerPoint: Slide-by-slide content extraction including speaker notes and embedded media.

Processing Pipeline

• Automatic chunking optimized per file type (paragraph-level for docs, slide-level for PPT).

• Vision model analysis for charts, diagrams, and screenshots.

• Metadata extraction (author, date, page count) stored alongside embeddings.

• Processing status webhooks for async file ingestion.