# Zilliz Cloud MCP > Zilliz Cloud MCP lets your agent manage and query vector data directly inside your AI workflow. You can list, create, drop, and describe entire collections in a cluster, all from natural language prompts. It executes high-performance vector similarity searches (ANN) while allowing you to filter results using complex metadata queries. Use it to power intelligent retrieval systems that understand context. ## Overview - **Category:** ai-frontier - **Price:** Free - **Tags:** similarity-search, vector-embeddings, ai-infrastructure, data-indexing, collection-management, milvus ## Description This MCP connects your AI agent straight into your Zilliz Cloud vector database. Instead of writing boilerplate code or navigating complex dashboards, you talk directly to your data structure. Your agent handles all the heavy lifting—managing collections, inserting new records, and running deep similarity searches using just language. For example, need to find documents related to 'Q3 financial compliance'? You tell your agent that, and it executes a precise search across your vector indexes. It even lets you refine those results by adding metadata filters, like only showing records from the 'Legal' department. If you're building complex data pipelines, connecting via Vinkius makes this MCP accessible to any compatible client, letting you treat your entire vector database management system as just another tool in your AI toolkit. ## Tools ### list_collections This tool retrieves a list of every collection currently stored in your Zilliz cluster. ### create_collection It creates a brand new vector collection within the cluster, requiring you to specify its structure. ### delete_entities This function removes specific data points or records from an existing collection. ### describe_collection It fetches detailed information, including the schema and status, for a single specified collection. ### drop_collection This tool permanently removes an entire vector collection from the cluster. ### insert_entities It adds new data, both vector and scalar, into a designated collection. ### load_collection This loads an entire specified collection into memory for immediate and optimized search availability. ### query_entities It searches for records using specific metadata filters (like dates or tags) to narrow down results. ### release_collection This function removes a collection from memory, freeing up cluster resources. ### search_vectors It performs the core task: executing a complex vector similarity search based on your configuration. ## Prompt Examples **Prompt:** ``` List all vector collections in my Zilliz cluster. ``` **Response:** ``` I've retrieved your collections. You have 3 collections: 'image_embeddings', 'text_docs', and 'product_features'. Would you like more details on any of them? ``` **Prompt:** ``` Show the schema and status for collection 'text_docs'. ``` **Response:** ``` Collection 'text_docs' is currently 'LOADED' with 150,000 rows. It has 3 fields: 'id' (Primary), 'vector' (FloatVector, 1536 dim), and 'metadata' (JSON). ``` **Prompt:** ``` Drop the collection named 'old_data_backup'. ``` **Response:** ``` Successfully dropped the collection 'old_data_backup'. All associated data has been permanently removed from your cluster. ``` ## Capabilities ### Inventorying and Defining Collections You can list all existing collections, describe a specific collection's schema, or create entirely new ones when setting up the infrastructure. ### Managing Data Lifecycle Your agent can load entire collections into memory for faster searching, release them when done, and delete outdated records from your vector indexes. ### Ingesting and Modifying Data The MCP allows you to insert new vector or scalar data into existing collections, or specifically delete entities that are no longer relevant. ### Performing Targeted Searches You execute complex vector similarity searches using customizable metrics. You can combine this with metadata filtering to pinpoint exact records within massive datasets. ## Use Cases ### Identifying Data Sources for a New Feature A developer needs to know what indexes exist before starting work. Instead of logging into the database console, they ask their agent to `list_collections`. The response instantly shows them all available vector collections, letting them start development immediately. ### Finding Specific Documents in a Large Archive A data scientist needs documents about 'Q2 marketing performance' but only from the 'West Coast' region. They ask their agent to `search_vectors` while simultaneously passing metadata filters, getting highly specific results without manual SQL joins. ### Preparing for a New Data Model An AI engineer wants to test a new data format. They use `create_collection` to build the schema and then run `insert_entities` with test vectors, all guided by conversational prompts. ### Cleaning Up Old Development Environments A team finishes testing an experimental dataset. Instead of manually deleting tables or running cleanup scripts, they prompt their agent to execute `drop_collection`, confirming that the old data is permanently gone. ## Benefits - Stop writing repetitive boilerplate code. Instead of manually calling `list_collections` to check available indexes, just ask your agent—it handles the inventory step for you. - Drastically improve search accuracy by combining vector similarity searches with metadata filtering. Use a prompt to run `query_entities` and narrow results down immediately. - Manage data flow efficiently. If you only need temporary access to a dataset, use `load_collection` when starting the task and remember to call `release_collection` when finished. - Maintain clean indexes by having your agent execute cleanup tasks like using `delete_entities` or running `drop_collection` on outdated data sets. - Run complex retrieval operations simply. Instead of writing search parameters, describe what you need and let the agent run `search_vectors` to find similar items. ## How It Works The bottom line is that it turns complex database operations into simple natural language commands executed by your agent. 1. First, subscribe to the Zilliz Cloud MCP and provide your specific Zilliz Cluster Endpoint and API Key. 2. Next, invoke the MCP through your AI client. You tell your agent exactly what you need—for instance, 'List all collections' or 'Search for similar images related to X'. 3. The MCP executes the necessary operations against your cluster and returns a clean, structured result set directly into your chat window. ## Frequently Asked Questions **How do I start using Zilliz Cloud MCP to find my collections?** You initiate the process by asking your agent to `list_collections`. This immediately returns a list of all available vector indexes in your cluster, letting you know where to focus. **Can I search for data using Zilliz Cloud MCP with filters?** Yes. You combine the powerful `search_vectors` tool with metadata filtering capabilities. This lets you narrow down billions of vectors to only those matching specific criteria, like date or department. **What is the best way to update data using Zilliz Cloud MCP?** You use `insert_entities` for adding new records. If a record needs correction or removal, you must explicitly tell your agent to run `delete_entities`. **Does the Zilliz Cloud MCP handle data cleanup?** Absolutely. You can use `drop_collection` to permanently eliminate old datasets, or use `release_collection` if you just need to temporarily free up memory resources without deleting the index. **Is Zilliz Cloud MCP only for reading data?** No. Beyond querying with `search_vectors`, this MCP allows full lifecycle management, including creating new indexes (`create_collection`) and inserting data (`insert_entities`). **How do I find my Cluster Endpoint?** You can find your Cluster Endpoint in the Zilliz Cloud Console under the 'Cluster Details' page. It typically looks like `https://in01-xxxxxxxxxxxx.vectordb.zillizcloud.com`. **Why do I need to 'load' a collection before searching?** Zilliz requires collections to be loaded into memory to perform high-performance similarity searches. Use the `load_collection` tool to make your data available for search. **Can I filter my vector search using metadata?** Yes, Zilliz supports hybrid search. You can use the `query_entities` tool for metadata-only filtering or include filtering expressions in your `search_vectors` JSON configuration.