IBA + MCP Live Simulation
Watch Intent-Bound Authorization protect Claude agents from prompt injection attacks
REAL-TIME DEMONSTRATIONSelect Attack Scenario
🔥 Moltbook Swarm Attack
Malicious agent posts injection prompts. IBA blocks propagation.
💰 Financial Data Exfiltration
Agent tries to access beyond declared "research" intent.
🏥 Medical Records Breach
Calendar agent pivots to healthcare data. IBA stops it.
Claude Agent Swarm (MCP)
IBA Validation Dashboard
Validations Processed
0
Attacks Blocked
0
Avg Validation Time
0ms
Protected Agents
0/5
Attack Flow Visualization
Event Log (Real-Time)
MCP + IBA Integration Code
# Install IBA wrapper for Anthropic MCP pip install iba-mcp-wrapper # Python implementation from iba import IntentDeclaration, IntentScope from iba.mcp import IBAMCPClient # Define intent for Claude agent intent = IntentDeclaration( intent_id="claude-research-001", declared_purpose="Research Q4 financial trends from public data", scope=IntentScope( allowed_resources=["web:search", "file:read:public"], forbidden_resources=["email:*", "database:customer"] ) ) # Wrap Claude MCP client with IBA validation client = IBAMCPClient(intent=intent) # All tool calls now validated against intent result = client.call_tool("search_web", {"query": "Q4 earnings"}) # ✓ ALLOWED - matches research intent result = client.call_tool("read_email", {"inbox": "CEO"}) # ✗ BLOCKED - violates forbidden resources