Vertex AI Vector Search MCP. Search embeddings and manage indices from chat.

You don't wanna jump into the Cloud Console just to run a vector search. This server plugs Google’s massive vector matching power right into your agent or IDE. It lets you manage billions of semantic embeddings conversationally. You treat it like infrastructure management for your AI client, not manual data entry.

If you need to find the closest thing in your dataset—the nearest neighbor—you use search_nearest_neighbors. All you gotta do is pass an endpoint ID, a deployed index ID, and the vector array generated by your query. The agent then takes those inputs and finds the absolute best semantic matches within the data you've indexed.

When you need to know what indexes exist across your whole Google Cloud project, list_vector_indexes gives you that complete rundown. You can check every single vector index configured, regardless of whether it’s active or not. If you want more detail on a specific index—like its dimensionality or how it's set up—you use get_index_details.

This pulls the full metadata for one named vector index so you know exactly what you're dealing with.

Your agent also needs to track where your services are running. To see all the service endpoints defined in your project, you call list_index_endpoints. These endpoints are critical because they scale your RAG applications and direct traffic. Once you have those endpoints listed, you can narrow it down by checking which specific vector indexes are actually deployed and active on a given endpoint using list_deployed_indexes.

That tells you if the index is ready for production reads.

And what about status? If you're working with multi-terabyte datasets, building or deploying an index isn't instantaneous. You need to track those long-running background operations. For that, use list_vector_operations to review the timeline and current status of tasks—like deployments or builds—so you know when your data is actually available for querying.

Basically, this server gives your agent total visibility into your vector search setup: it lets you list every index (list_vector_indexes), check its config (get_index_details), see which network endpoints are active (list_index_endpoints), confirm what indexes are running on those endpoints (list_deployed_indexes), monitor if the builds finished (list_vector_operations), and finally, run the search itself (search_nearest_neighbors).

It’s a complete toolset for managing vector databases without leaving your chat window.