Voyage AI MCP. Get high-precision vectors from text, images, and code.

You're building a sophisticated RAG system, and you need reliable vector generation that handles everything—plain text, images, whole documents. This server gives your agent high-precision embeddings and reranking capabilities right out of the gate. It’s built for deep context retrieval.

Generating Vectors and Context

You can start by creating standard embeddings using create_embeddings. You feed it simple text strings, and it spits out high-dimensional vectors that you use in your database. If your data includes images alongside text, don't worry; you'll generate combined vectors using create_multimodal_embeddings, which lets your agent run visual searches just as easily as text queries.

For documents, you're gonna want more than basic embeddings. Use create_contextualized_embeddings when you process document chunks. This tool embeds the chunk while keeping track of the surrounding context and metadata; that means when you retrieve it later, you don't lose vital information about where in the document the snippet came from.

If you just run with standard embeddings on large documents, you risk losing this critical contextual depth.

Refining Search Results

Initial search results are only half the battle. You gotta make sure the top hits are actually the best ones. Use rerank to take a list of initial document chunks and reorder them based on how relevant they really are to your specific query. This process, which uses cross-encoders, makes certain that when your agent reads the context, it's reading the most accurate info first.

Handling Massive Data Loads (Batch Processing)

When you’re dealing with thousands of files—say, an entire corporate knowledge base—you can't process them all at once. You start by uploading those assets using upload_file, which designates multiple files for later batch work. To kick off the large-scale transformation, you call create_batch. This submits a job to process your whole dataset asynchronously.

You need to keep an eye on that process. Use list_batches to see every job currently running or waiting in line. If you want to check the progress of one specific job, use get_batch, which tells you exactly where it stands. If a batch job goes sideways or you change your mind, you can shut it down instantly using cancel_batch, provided you have its unique ID.

Managing Your Files and Assets

The server tracks everything you upload. To see what files are sitting there waiting for processing, just run list_files. If you need details on a specific file—like its size or file type—use get_file to grab that metadata. Need the raw content of an uploaded file? You'll download it using get_file_content.

And when you’re done with a file and want to clean up your storage, use delete_file to remove it from the server’s tracked space.

This suite handles everything: embedding creation for text or images, making sure context stays intact, running massive jobs in the background, and fine-tuning search results until they're perfect.