# Voyage AI MCP for AI Agents AI Agent Connect > Voyage AI (AI Embeddings API) is an Connector that lets your agent generate high-quality text, multimodal, and contextualized embeddings. It also provides high-precision reranking for RAG workflows, making it easier to find relevant data in large datasets. ## Overview - **Category:** developer-tools - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_x7lwFbbAdz5StvxoOu2Tkh9eDMu9Uh5i3GKObLMP/ai-agent-connect - **Tags:** embeddings, rag, rerank, vector-search, multimodal-ai ## Description This Connector connects your AI agent to Voyage AI's embedding and reranking models. If you're building a search system or a RAG pipeline, you know that standard embeddings sometimes miss the mark because they lose the surrounding context. This tool fixes that by letting your agent create embeddings that actually respect the flow of the document. You can also feed it images and text together for multimodal search or use its reranking capabilities to double-check the accuracy of your search results before showing them to a user. It's designed for people who need production-grade retrieval where close enough isn't good enough. You can manage these operations through the Vinkius catalog, which makes it easy to plug into your existing agent setup without worrying about the underlying infrastructure. ## Tools ### cancel_batch Stop a batch job if it's no longer needed. This saves on costs and resources when you need to pivot quickly. ### create_batch Start a background job for large-scale data processing. This is the best way to handle thousands of items at once. ### create_contextualized_embeddings Create embeddings that keep the surrounding context of a document chunk. This significantly reduces errors in long-form text retrieval. ### create_embeddings Generate vectors for text or code. It's the standard way to turn your data into numbers for search. ### create_multimodal_embeddings Turn images and text into a single vector space. Use this for visual search across different types of media. ### delete_file Remove a specific file from your storage. This helps keep your data clean during batch processing. ### get_batch Check the status of a background job. Use this to see if your large-scale embedding task is finished. ### get_file_content Download the content of a specific file. This lets your agent read data directly from your storage. ### get_file Retrieve metadata about a file. Use this to check file details without downloading the whole thing. ### list_batches See a list of all your current and past batch jobs. It's great for monitoring high-volume data updates. ### rerank Take a list of search results and reorder them by relevance. It ensures your agent sees the most important info first. ### upload_file Send a file to the system for batch inference. This is how you start processing large amounts of data at once. ### list_files View all files in your storage. Use this to see what's available for your agent to process. ## Prompt Examples **Prompt:** ``` I have a long legal contract. How can I make sure my agent understands the context of specific clauses? ``` **Response:** ``` To ensure your agent understands the context, you should use contextualized embeddings. Instead of just grabbing a sentence, this method looks at the surrounding text to keep the meaning intact. This is great for legal docs where a single line can change based on the paragraph above it. ``` **Prompt:** ``` Can I search for products using both photos and text descriptions? ``` **Response:** ``` Yes, you can. By using multimodal embeddings, your agent can map both images and text into the same vector space. This means a user can upload a photo of a lamp and your agent can find it alongside text searches for 'modern table lamps'. ``` **Prompt:** ``` I need to vectorize 5,000 product descriptions. What's the best way to do this? ``` **Response:** ``` For that volume, you should use the batch processing tools. You can upload your file and start a batch job. This handles the 5,000 items in the background so you don't have to worry about individual request limits or manual monitoring. ``` ## Capabilities ### Generate text and code vectors Use create_embeddings to turn your text or code into numbers for search. ### Create multimodal vectors Use create_multimodal_embeddings to map images and text into a single vector space. ### Maintain document context Use create_contextualized_embeddings to ensure chunks keep the meaning of the surrounding text. ### Reorder search results Use rerank to ensure your agent sees the most relevant information at the top of the list. ### Handle large-scale data Use create_batch to process thousands of items in the background without hitting rate limits. ## Use Cases ### Improving RAG accuracy for long documents A user asks about a specific clause in a 100-page contract. The agent uses create_contextualized_embeddings to ensure it doesn't lose the surrounding context. ### Visual search for e-commerce A shopper uploads a photo of a chair. The agent uses create_multimodal_embeddings to find similar items in the product catalog. ### Processing massive datasets A data scientist needs to vectorize 1 million product descriptions. They use upload_file and create_batch to process the data without hitting rate limits. ### Refining search results for a help desk A support agent gets 20 similar articles. The agent uses rerank to put the most helpful one at the very top. ## Benefits - Better retrieval accuracy because create_contextualized_embeddings keeps the surrounding context of your data intact. - Faster large-scale data processing by using create_batch to handle thousands of files in the background. - More versatile search capabilities with create_multimodal_embeddings for both images and text. - Higher precision results by using rerank to ensure the most relevant content reaches your agent first. - Easier management of your data assets with tools like list_files and list_batches to keep everything organized. ## How It Works The bottom line is you get production-ready search results without managing the complex math behind the embeddings. 1. Subscribe to this Connector and enter your Voyage AI API Key. 2. Tell your agent which model to use for your specific data type, like text, code, or images. 3. Get back high-dimensional vectors or reranked lists ready for your database. ## Frequently Asked Questions **How does Voyage AI MCP improve my RAG system's accuracy?** It provides contextualized embeddings that keep the surrounding information of a document chunk in mind. This prevents the AI from getting confused by isolated sentences and leads to much more relevant answers. **Can I use Voyage AI MCP to search through images and text together?** Yes, the multimodal capabilities allow your agent to search across different types of media in one go. This is perfect for visual search or catalogs with mixed content. **Is Voyage AI MCP good for handling very large datasets?** Yes, it includes dedicated batch tools. You can upload files for background processing, which is the standard way to handle thousands of items without hitting rate limits. **What is the difference between standard embeddings and Voyage AI's contextualized ones?** Standard embeddings treat every chunk as a standalone island. Voyage AI's contextualized versions look at the surrounding text to ensure the meaning stays consistent, which is vital for long-form content. **How does reranking help my AI agent's responses?** Reranking takes the top results from a search and re-scores them for relevance. It ensures that the absolute best match is moved to the top, so your agent always sees the most important information first. **Can I manage my batch jobs using the Voyage AI MCP?** Yes, you can list all your batches and check the status of specific jobs. This makes it easy to monitor large data updates as they happen in the background. **How does reranking improve my RAG system's accuracy?** By using the `rerank` tool, your agent can take a list of potentially relevant documents and re-score them using a powerful cross-encoder model. This ensures that the most semantically relevant pieces of information are ranked first, providing better context for the LLM to answer queries. **What is the benefit of using contextualized embeddings?** The `create_contextualized_embeddings` tool allows you to embed chunks of text while considering the surrounding content of the same document. This prevents loss of meaning that often happens with standard chunking, leading to much higher retrieval precision. **Can I process images and text in the same vector space?** Yes! With `create_multimodal_embeddings`, you can provide interleaved sequences of text and image URLs. Voyage AI will generate a single embedding that represents the combined semantic meaning, perfect for visual or hybrid search.