mcp-trust-harness

sohepalslamat/mcp-trust-harness

★ 0 stars Python AI/LLM Updated today

Proof-of-concept harness for runtime trust evaluation and just-in-time consent when an AI model discovers an unknown MCP server.

View on GitHub → Try with Claude — $10 free →

Quick Install

Copy the config for your editor. Some servers may need additional setup — check the README.

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-trust-harness": {
      "command": "uvx",
      "args": [
        "mcp-trust-harness"
      ]
    }
  }
}

Or install with pip: pip install mcp-trust-harness

README Excerpt

A proof-of-concept **harness** + **48-case decision benchmark** that give an AI model the capability it does *not* have today: to **discover a previously-unknown MCP server at runtime, evaluate its trust, decide whether to connect, obtain just-in-time user consent, act through an ephemeral connection, and disconnect** —

Tools (1)

refuse

mcp-trust-harness

Quick Install

README Excerpt

Tools (1)

Related AI/LLM Servers