Show HN: Building High-Performance AI Agents with SmartBuckets and MCP
## The Problem
Building knowledge-powered AI agents typically takes 6+ months of engineering work. Teams spend months on:
- Document processing pipelines - Chunking strategies - Embedding generation - Entity extraction and knowledge graph creation - Vector database configuration - Retrieval algorithm development - Context assembly and management
Our Solution
SmartBuckets eliminates the need to build these components from scratch, providing a complete knowledge engine that integrates with MCP for direct model access. The technical architecture looks like this:
AI Decomposition
When you upload a file to a SmartBucket, it triggers an intelligent process we call AI decomposition. This process is fundamental to understanding how SmartBuckets transform raw files into AI-enhanced resources. Let’s look at what happens when you upload a PDF....
The decomposition process works in several stages:
1. First, the system identifies and extracts different types of content from your file - text, images, tables, metadata and more. 2. Each component is then processed through specialized AI models designed for that specific type of content 3. The enhanced data is stored in optimized datastores, maintaining relationships between different components 4. All of this processed information becomes immediately available for AI queries
Automatic Knowledge Graph Creation
What sets SmartBuckets apart is its more than just vector search, it also has automatic knowledge graph capabilities. When you upload documents, the system:
1. Automatically extracts entities and relationships 2. Constructs a knowledge graph connecting related information 3. Enriches data with metadata for improved retrieval
These knowledge graphs significantly reduce model hallucinations and improve recall of relevant information.
AI models and AI data stores
When you upload data to a SmartBucket, our AI pipeline analyzes it and stores the results in multiple specialized systems including vector stores, graph databases, and relationship stores.
The processing pipeline includes several analysis models that:
- Detect PII (Personal Identifiable Information) - Screen for harmful content (coming soon) - Much more that won't fit in the 4000 character HN limit
Technical Implementation Example
Adding SmartBuckets to MCP-compatible systems requires minimal code. If you wanted to attach it to Claude Desktop:
1. Claude - Settings - Developer - Edit Config 2. Simply add this code snippet (using your API key from your http://liquidmetal.run account under → Settings → API Keys
```json
json { "mcpServers": { "liquidmetal": { "command": "npx", "args": [ "mcp-remote", "https://mcp.raindrop.run/sse", "--header", "Authorization: Bearer ${RAINDROP_API_KEY}" ], "env": { "RAINDROP_API_KEY": "<LIQUIDMETAL_KEY_HERE>" } } } }
```
1. Start using your documents in conversations right in Claude Desktop or anything MCP.
What's Next
We're working on:
- Direct SmartBucket CREATE feature via MCP - Video, Code, Logs, and more expanded file type support - Whatever you ask for… so please let us know what is missing.
We're releasing this integration now and would love the HN community's feedback. Try it out at https://docs.liquidmetal.ai/ and use this code: HN-MCP-100 to get $100 in free LiquidMetal credits.