Live Demonstration
IBA + MCP Integration
Watch the boundary hold.
Illustrative scenarios showing how an IBA wrapper around Anthropic’s Model Context Protocol would block out-of-scope tool calls. Real, deterministic logic below — not a claim of a deployed production system.
0Validations
0Blocked
0msLast Response
Ready. Select a scenario above.
Integration
MCP + IBA Pattern
# Declare intent — the only thing the agent is authorised to do
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 the MCP client — all tool calls now validated against intent
client = IBAMCPClient(intent=intent)
result = client.call_tool(“search_web”, {“query”: “Q4 earnings”})
# -> ALLOWED, matches declared research intent
result = client.call_tool(“read_email”, {“inbox”: “CEO”})
# -> BLOCKED, violates forbidden_resources list
This is a real, runnable pattern — see the verified iba-x-demo repository for a fully tested working implementation.