openai-agents-trust-gate

Trust-gated MCP server wrapper for OpenAI Agents SDK. Wraps any MCPServer to check behavioral trust scores via Dominion Observatory before every tool call.

Install

pip install openai-agents-trust-gate

Quick Start

from agents.mcp import MCPServerStreamableHttp
from openai_agents_trust_gate import TrustGatedMCPServer

raw_server = MCPServerStreamableHttp(url="https://example.com/mcp")
server = TrustGatedMCPServer(raw_server, min_score=70)

agent = Agent(
    name="my-agent",
    mcp_servers=[server],  # Trust-gated automatically
)

How It Works

1. Wraps any MCPServer, intercepting call_tool() at the MCP layer
2. Before each tool call, checks the server trust score via Dominion Observatory
3. Blocks calls to servers below the threshold
4. Reports outcomes back to the Observatory
5. Scores are cached (5-minute TTL) for performance

Configuration

TrustGatedMCPServer(
    inner_server,
    min_score=70,           # Minimum trust score (0-100)
    cache_ttl=300.0,        # Cache TTL in seconds
    block_on_error=False,   # Fail-open if Observatory unreachable
    report_outcomes=True,   # Report call outcomes back
)

License

MIT