Capsa

What it does\nCapsa provides a lightweight system for indexing and retrieving documents using semantic embeddings. It connects to an OpenAI-compatible embedding service and stores data in a libSQL database, allowing AI agents to perform semantic searches across PDFs, YouTube transcripts, and plain text.\n\n## Tools\n- pdf_index: Ingests PDF files, extracts text and metadata, and stores them as embeddings.\n- youtube_index: Downloads YouTube transcripts and indexes them for semantic retrieval.\n- semantic_search: Performs natural language queries against the stored document vectors using cosine similarity.\n\n## Installation\nAdd to claude_desktop_config.json (requires a running embedding server like llama.cpp on port 9000):\njson\n{\n \"mcpServers\": {\n \"capsa\": {\n \"command\": \"/absolute/path/to/capsa/target/release/capsa\",\n \"env\": { \"CAPSA_API_KEY\": \"your_key_here\" }\n }\n }\n}\n\n\n## Supported hosts\n- claude