✨ Key Features
Native MCP
First-class Model Context Protocol support — connect to any MCP server directly
Advanced Reasoning
Powered by Claude's state-of-the-art reasoning and analysis capabilities
A2A Adapt
Generate an A2A 1.0 Agent Card and server from an existing Claude SDK agent
GEPA Optimization
Optimize system prompts and tool descriptions with SuperOptiX GEPA
Discovery Mode
Dynamic tool discovery for large tool catalogs (100+ tools)
Streaming
Async streaming responses for real-time agent interactions
📦 Installation
Install SuperOptiX with Claude SDK
uv tool install superoptix
Set Anthropic API Key
export ANTHROPIC_API_KEY="sk-ant-..."
Recommended Models:
- •
claude-sonnet-4-5— Balanced performance (recommended) - •
claude-opus-4-5— Maximum capability - •
claude-haiku-4-5— Fastest, cost-effective
🚀 Quick Start
Get started with Claude Agent SDK in minutes
Pull Demo Agent
super agent pull assistant_claude
Compile
super agent compile assistant_claude --framework claude-sdk
Evaluate
super agent evaluate assistant_claude
Optimize
super agent compile assistant_claude --framework claude-sdk --optimize super agent optimize assistant_claude --auto light --framework claude-sdk
Run
super agent run assistant_claude --framework claude-sdk --goal "What is AI?"
⚙️ Model Configuration
Claude Sonnet 4.5
Balanced performance and cost
language_model: location: cloud provider: anthropic model: claude-sonnet-4-5 temperature: 0.2
Claude Opus 4.5
Most capable, complex reasoning
language_model: location: cloud provider: anthropic model: claude-opus-4-5 temperature: 0.2
Claude Haiku 4.5
Fastest, ideal for simple tasks
language_model: location: cloud provider: anthropic model: claude-haiku-4-5 temperature: 0.2
🎯 What GEPA Optimizes
Primary Target: System Prompt
GEPA optimizes the system_prompt field — the core instruction guiding agent behavior with Claude.
Expected Improvement
Example Optimization Transformation
Before Optimization:
You are a helpful assistant.After GEPA:
You are an expert AI assistant specialized in providing accurate, well-structured responses.
When answering:
1. Analyze the question carefully
2. Provide factual, sourced information
3. Use clear formatting with examples
4. Acknowledge uncertainty when appropriate📋 SuperSpec Playbook
apiVersion: agent/v1 kind: AgentSpec metadata: name: assistant_claude id: assistant_claude namespace: demo version: 1.0.0 spec: target_framework: claude-sdk language_model: location: cloud provider: anthropic model: claude-sonnet-4-5 temperature: 0.2 persona: role: AI Assistant goal: Provide helpful, accurate responses instructions: | You are a knowledgeable AI assistant powered by Claude. Provide clear, well-structured answers. tasks: - name: answer_question instruction: "Answer the user's question accurately." inputs: - name: question type: str required: true outputs: - name: answer type: str feature_specifications: scenarios: - name: Basic question input: question: "What is machine learning?" expected_output: expected_keywords: - machine - learning - algorithm
Claude SDK vs Other Frameworks
| Feature | Claude SDK | OpenAI SDK | DSPy |
|---|---|---|---|
| Native MCP | Native | ❌ No | ❌ No |
| Reasoning | Claude 4.5 | GPT-4o | Any LLM |
| Ollama Support | ❌ Cloud only | Yes | Yes |
| Streaming | Async | Yes | Limited |
| GEPA Optimization | system_prompt | instructions | 10+ variables |
Best for:
- MCP-powered tool integrations
- Claude's advanced reasoning tasks
- A2A 1.0 endpoints for existing agents
- Streaming agent interactions
- Production cloud-based deployments
Consider Other Frameworks:
DSPy
If you need maximum optimization (10+ variables) or local Ollama models
OpenAI SDK
If you want Ollama support and simple handoffs
CrewAI
If you need multi-agent team collaboration