Forge

Connects Claude and other MCP clients to Forge, Voxell's hosted text-embedding API running the Qwen3-Embedding family. You get two tools: embed text into vectors (with input_type flags for query vs. document, Matryoshka dimension truncation, and three quality tiers) and list_models to see what's available. The ultra tier runs the 8B model that sits at #4 on MTEB English. Voxell doesn't store your text or vectors, just token counts for billing. Useful when you need semantic search, RAG retrieval, or similarity scoring without managing your own embedding infrastructure. Also speaks OpenAI's embeddings API if you want to swap it into existing code that already calls text-embedding-3-large.