mcpdirectory.dev
9,129 servers · updated 23 Sept

Protect MCP server

official from Scopeblind Active last commit 3 mo
{
  "mcpServers": {
    "scopeblind-protect": {
      "command": "npx",
      "args": [
        "-y",
        "protect-mcp@latest"
      ],
      "env": {
        "SCOPEBLIND_TOKEN": "<scopeblind_token>"
      }
    }
  }
}
file .mcp.json (in your project root)install npx -y protect-mcpversion 0.13.1

What it does

A Cedar policy gate that evaluates MCP tool calls against rules and signs Ed25519 receipts of each decision. Designed for developers who want to control what tools an AI agent can call, with cryptographic proof of decisions and offline verifiability.

  • Wrap an MCP server to log all tool calls and inspect them in a local dashboard before enforcing a policy.
  • Define Cedar policies that allow, block, or require approval for specific tool calls, with fail-closed denial on any policy error.
  • Generate signed receipts of tool call decisions that can be verified offline without trusting the server, for audit trails.

Tools this server exposes

ToolDoes
evaluate_actionEvaluate a proposed tool call against a Cedar policy and return whether it is allowed, denied, or requires approval.
sign_decisionTurn a tool call decision into an Ed25519-signed receipt that can be verified offline.
verify_receiptVerify an Ed25519-signed receipt against a public key to confirm a decision was not tampered with.
self_testRun a startup proof that the gate denies a known-forbidden action and that receipts verify correctly.

Before you enable it

An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.