ScopeBlind Examples

April 8, 2026 ยท View on GitHub

License: MIT IETF Draft

ScopeBlind Examples

Complete examples showing how to add signed, independently verifiable audit trails to AI agent tool calls, including cybersecurity use cases for vulnerability disclosure.

Each example uses protect-mcp to wrap MCP tool servers with Cedar policies and Ed25519-signed receipts. Every tool call produces a cryptographic receipt that can be verified offline by anyone -- without contacting the original issuer.

Examples

ExampleDescriptionTime
claude-code-hooksAdd protect-mcp as Claude Code hooks. Every tool call gets a signed receipt and Cedar policy check.~2 min
express-api-gatewayWrap an Express-based MCP server with JSON policies and rate limiting.~5 min
mcp-server-signingCedar WASM policy engine with per-tool authorization and full audit bundles.~10 min
takt-workflow-receiptsAdd signed receipts to TAKT multi-step workflows. Level 1 external integration.~5 min
security-vulnerability-disclosureReceipt-signed vulnerability disclosure lifecycle with Cedar governance policies. For AI security agents.~10 min

Prerequisites

  • Node.js 20+
  • npm 9+

No ScopeBlind account required. All examples run locally.

How verification works

Every tool call through protect-mcp produces a signed receipt containing:

  1. Decision -- whether the call was allowed, denied, or logged (shadow mode)
  2. Policy hash -- SHA-256 of the Cedar or JSON policy that produced the decision
  3. Timestamp -- when the decision was made
  4. Tool context -- tool name, truncated input hash, trust tier
  5. Ed25519 signature -- signs all of the above

Receipts are appended to .protect-mcp-receipts.jsonl (one JSON object per line). Anyone can verify them offline using:

npx @veritasacta/verify .protect-mcp-receipts.jsonl

This checks every signature without contacting any server. The verifier is issuer-blind -- it validates cryptographic integrity without knowing or trusting the original signer.

Ecosystem