Skip to content

CLAUDE_INTEGRATION.md

Latest commit

 

History

History
101 lines (76 loc) · 2.19 KB

CLAUDE_INTEGRATION.md

File metadata and controls

101 lines (76 loc) · 2.19 KB
Raw

Claude Integration Example

This example shows how to integrate GhidraInsight with Claude API for AI-powered binary analysis.

Setup

1. Install Dependencies

pip install anthropic ghidrainsight

2. Set API Keys

export CLAUDE_API_KEY="sk-ant-..."
export GHIDRA_API_KEY="your-ghidra-api-key"

3. Run the Example

python claude_integration.py --binary /path/to/binary --question "What vulnerabilities exist?"

How It Works

  1. Upload Binary: Sends binary to GhidraInsight for analysis
  2. Get Results: Retrieves analysis results (crypto, vulnerabilities, taint)
  3. Build Context: Creates a detailed context for Claude
  4. Query Claude: Sends question + context to Claude API
  5. Stream Response: Streams Claude's response in real-time

Example Usage

Basic Analysis

from claude_integration import BinaryAnalyzer

analyzer = BinaryAnalyzer("http://localhost:8000")
result = analyzer.analyze(
    binary_path="binary.elf",
    question="What cryptographic algorithms are used?"
)
print(result)

Advanced Queries

queries = [
    "Identify all potential buffer overflow vulnerabilities",
    "Which functions handle user input and where does it flow?",
    "What cryptographic weaknesses exist?",
    "Suggest hardening recommendations",
]

for query in queries:
    result = analyzer.analyze("binary.elf", query)
    print(f"Q: {query}")
    print(f"A: {result}\n")

Streaming Output

for chunk in analyzer.analyze_stream("binary.elf", "What are the main security issues?"):
    print(chunk, end="", flush=True)

Features

  • ✅ Real-time binary analysis with GhidraInsight
  • ✅ Claude API integration
  • ✅ Multi-turn conversation support
  • ✅ JSON output for integration
  • ✅ Error handling and retries
  • ✅ Streaming responses

Output Format

{
  "query": "What vulnerabilities exist?",
  "analysis": {
    "crypto": [...],
    "vulnerabilities": [...],
    "taint": [...]
  },
  "response": "Based on the analysis, I found...",
  "tokens_used": {
    "input": 2048,
    "output": 512
  }
}

Note: Requires Claude API access. Get your API key from https://console.anthropic.com