Skip to main content

Weaviate Vector Database: 6 Tools for AI-Native Applications

· 8 min read
Tony Lewis
MCPBundles

Weaviate is an open-source vector database that powers AI-native applications—RAG systems, semantic search, recommendation engines, and more. But how do you make a vector database accessible to AI agents? You can't just expose raw API endpoints and expect good results.

The answer: 6 focused tools organized around what developers actually do with vector databases. Not 20 tools covering every edge case. Not 3 tools that force you into awkward patterns. Just 6 tools that handle search, storage, browsing, and management—the core workflows every vector database application needs.

The Tool Architecture​

Here's how we organized Weaviate into 6 focused tools:

Weaviate Tools Overview

Use Cases → Tools​

"I need to search my documents semantically"​

Use Case: Semantic search and retrieval

Tools: search, fetch

When you're building RAG applications or document search, you need to find relevant content based on meaning, not just keywords. This workflow gives you:

  • Semantic search: Natural language queries that find relevant content by meaning
  • Hybrid search: Combines semantic vectors with keyword matching for better results
  • Smart retrieval: Fetch full documents or objects using smart IDs

Example workflow: Search "customer support policies" → Get ranked results with relevance scores → Fetch full document details → Use in RAG context

How it works:

  • search performs hybrid semantic + keyword search across your collections
  • Returns ranked results with IDs, titles, and relevance scores
  • fetch retrieves complete objects using smart IDs like weaviate:object:Documents:uuid

"I want to store and retrieve data from my vector database"​

Use Case: Data storage and retrieval

Tools: weaviate_upsert, fetch

Storing data in a vector database requires inserting objects and being able to retrieve them later. This workflow provides:

  • Unified insert/update: One tool handles both inserts and updates, single items or batches
  • Smart routing: Fetch any object, schema, or collection metadata using smart IDs
  • Flexible operations: Insert new objects or update existing ones by providing IDs

Example workflow: Insert product documents → Update pricing information → Fetch specific product → Use in application

How it works:

  • weaviate_upsert accepts data as arrays—[obj] for single, [obj1, obj2] for batch
  • Provide ids array for updates, omit for inserts
  • fetch uses smart IDs: weaviate:object:Products:uuid for objects, weaviate:schema:Products for schemas

"I need to browse and explore my vector database"​

Use Case: Data exploration and discovery

Tools: weaviate_list_collections, weaviate_list_objects

Understanding what's in your vector database is essential for building applications. This workflow enables:

  • Collection discovery: See all collections and their schemas
  • Object browsing: Explore objects with filtering, sorting, and pagination
  • Data inspection: Understand your data structure and content

Example workflow: List all collections → See collection schemas → Browse objects in a collection → Filter by property → Sort and paginate results

How it works:

  • weaviate_list_collections shows all collections with optional schema and object counts
  • weaviate_list_objects provides granular control: search filtering, WHERE conditions, property selection, sorting, pagination
  • Both tools include rich metadata to help you understand your data

"I need to manage my vector database collections"​

Use Case: Database administration and maintenance

Tools: weaviate_list_collections, weaviate_delete

Managing collections and cleaning up data is part of maintaining a healthy vector database. This workflow helps you:

  • Collection management: See all collections, their schemas, and object counts
  • Data cleanup: Delete individual objects, bulk delete, or remove entire collections
  • Database inspection: Understand collection structure and data distribution

Example workflow: List collections → Check object counts → Delete outdated objects → Remove unused collections

How it works:

  • weaviate_list_collections provides collection metadata and schema information
  • weaviate_delete supports three modes: single object, bulk objects, or entire collection removal
  • Safety warnings for destructive operations (collection deletion is permanent)

Why Only 6 Tools?​

We started with 10 tools covering every edge case. Then we realized: more tools doesn't mean better AI agent usage. Here's what we learned:

The Problem with 10 Tools​

Initial design had separate tools for:

  • weaviate_insert_one vs weaviate_batch_insert vs weaviate_update_object
  • weaviate_get_schema separate from object fetching
  • weaviate_filter redundant with list_objects WHERE parameter

Result: AI agents had to choose between similar tools, leading to confusion and wrong tool selection.

The Solution: 6 Tools​

Consolidated operations:

  • weaviate_upsert handles all insert/update operations (single or batch)
  • fetch with smart ID routing handles objects, schemas, and collections

Rich parameters:

  • weaviate_list_objects has 8 parameters for granular control
  • Parameter descriptions explain not just what but when and why

Result: 40% fewer tools, zero functionality loss, clearer mental model.

The Mental Model​

2 Universal Tools (work with any provider):

  • search - Semantic search following OpenAI MCP standard
  • fetch - Smart ID routing for any resource type

4 Domain Tools (Weaviate-specific):

  • weaviate_list_collections - Collection discovery
  • weaviate_list_objects - Object browsing
  • weaviate_upsert - Data storage
  • weaviate_delete - Data removal

This clear separation helps AI agents understand: use universal tools for discovery, domain tools for operations.

How This Improves AI Agent Usage​

When an AI agent connects to Weaviate through MCPBundles, it can now:

"Find documents about authentication" → Uses search for semantic search → Gets ranked results with relevance scores → Fetches full documents using fetch

"Store these product descriptions in my vector database" → Uses weaviate_upsert with data array → Automatically handles batch insertion → Returns IDs for later retrieval

"Show me what's in my Documents collection" → Uses weaviate_list_objects with collection name → Applies filtering and sorting → Returns paginated results

Instead of choosing between 10 similar tools, the AI uses the right tool for the workflow. This follows our bundle design philosophy where we organize tools around workflows, not API endpoints.

Smart Design Patterns​

Smart ID Routing​

Instead of separate tools for fetching objects vs schemas vs collections, we use one fetch tool with smart IDs:

  • weaviate:object:Products:uuid - Fetch a specific object
  • weaviate:schema:Products - Get collection schema
  • weaviate:collections:list - List all collections

The AI learns: "I can fetch different resource types by changing the ID format."

Unified Upsert​

Instead of separate insert/update/batch tools, one weaviate_upsert tool handles everything:

  • Insert mode: Omit ids parameter → creates new objects
  • Update mode: Provide ids array → updates existing objects
  • Batch mode: Always use array format, even for single items

The AI learns: "Array format is always required, and IDs determine create vs update."

Rich Parameter Descriptions​

Every parameter description explains not just what it is, but when to use it and why it matters:

Bad: "Maximum number of results"

Good: "Maximum number of results to return. Range: 1-100. Higher limits may impact performance but provide more comprehensive results. Default is 10, which is optimal for most use cases."

This approach follows our tool parameter design principles where descriptions teach AI agents how to use tools effectively.

Real-World Use Cases​

RAG Application​

Workflow: Store documents → Search semantically → Retrieve context

  1. Use weaviate_upsert to store documents with embeddings
  2. Use search to find relevant documents for queries
  3. Use fetch to get full document content for RAG context

Recommendation Engine​

Workflow: Store user preferences → Search similar items → Return recommendations

  1. Use weaviate_upsert to store user profiles and item data
  2. Use search to find similar items based on user preferences
  3. Use weaviate_list_objects to browse and filter catalog

Workflow: Index documents → Search by meaning → Browse results

  1. Use weaviate_upsert to index documents
  2. Use search for semantic search queries
  3. Use weaviate_list_objects to browse and filter search results

Frequently Asked Questions​

Q: Do I need all 6 tools?

No. Most applications use search, fetch, and weaviate_upsert for core workflows. weaviate_list_collections and weaviate_list_objects are for exploration and administration. weaviate_delete is for cleanup operations.

Q: Can I use Weaviate without vector embeddings?

Yes. Weaviate supports keyword-only search (BM25) if no vectorizer is configured. The search tool automatically falls back to BM25 when vectors aren't available.

Q: How do I know which collection to use?

Use weaviate_list_collections to see all available collections and their schemas. Each collection has its own schema defining properties and data types.

Q: What's the difference between search and list_objects?

search performs semantic/keyword search across collections and returns ranked results. weaviate_list_objects browses objects in a specific collection with filtering and sorting. Use search for discovery, list_objects for browsing.

Q: Can I update objects without knowing their IDs?

No. Updates require object IDs. Use search or weaviate_list_objects to find objects first, then use their IDs with weaviate_upsert for updates.

Try It Yourself​

Weaviate is now available on MCPBundles with all 6 tools ready to use:

  1. Connect your Weaviate instance (instance URL and API key)
  2. Enable the Weaviate bundle
  3. Start building RAG applications with semantic search

The AI will automatically use the right tools based on your intent—search for semantic retrieval, upsert for storage, fetch for retrieval.

Want to see this in action? Try Weaviate semantic search free and build your first RAG application with AI-powered document retrieval.

The future of vector databases isn't exposing every API endpoint—it's organizing tools around workflows that AI agents understand. That's what our 6-tool design enables.