ToolTrust

ToolTrust Scanner for MCP Servers and AI Agents

AI Agent Tool Security Directory

Powered by ToolTrust Scanner, this live directory helps you review prompt injection, permission risk, supply-chain issues, and scope mismatches before your agent trusts a tool.

🚨 Prompt Injection⚠️ Excess Permissions💻 Code Execution🚨 Supply Chain🎭 Typosquatting👥 Tool Shadowing

1233

Tools Scanned

579

Safe (S/A/B)

588

Medium Risk (C)

66

Risky (D/F)

Scan your MCP servers

Via MCP (recommended) — works inside Claude Code, Cursor, Claude Desktop

{
  "mcpServers": {
    "tooltrust": {
      "command": "npx",
      "args": ["-y", "tooltrust-mcp"]
    }
  }
}

Then ask your agent: run tooltrust_scan_config

Via CLI

$ curl -sfL https://raw.githubusercontent.com/AgentSafe-AI/tooltrust-scanner/main/install.sh | bash
$ tooltrust-scanner scan --server "npx -y @modelcontextprotocol/server-filesystem /tmp"
Full docs & GitHub Actions integration →

What the decisions mean

ToolTrust maps findings to three actions so you know what to do next after a scan.

Allow

The server is low risk enough to use normally. You can enable it in day-to-day workflows, though it is still worth reviewing findings before broad deployment.

Require approval

Keep the server behind manual review. Do not run it unattended until you review the findings, keep a human in the loop for risky tool calls, and reduce permissions before enabling automation.

Block

Do not use this server in production. If a tool is graded D or F, disable it in your production .mcp.json while waiting for a fix and re-scan.

{
  "mcpServers": {
    "tool-name": {
      "disabled": true
    }
  }
}