Configure Cedar policy enforcement and Ed25519 signed receipts for Claude Code tool calls. Use when setting up projects that need cryptographic audit trails, policy-gated tool execution, or compliance-ready evidence of agent actions.
Detected risks:
protect-mcp is a governance and auditability skill for Claude Code projects that require policy-controlled tool execution and cryptographic evidence of agent actions. It addresses key gaps in standard tool usage by enforcing authorization decisions before tools run and generating tamper-evident records after execution. The skill is designed for environments where audit trails, integrity verification, and compliance evidence are important. By combining Cedar policy evaluation with Ed25519-signed receipts, it helps create verifiable records of tool decisions that can be independently validated without relying on the original operator or system.
The skill evaluates every tool invocation against a Cedar policy before execution, with deny decisions preventing the tool from running. After execution, it generates signed receipts that capture information such as tool inputs, outputs, policy-related decision data, and timestamps. Receipts are hash-chained and can be verified offline using the provided verification tooling. Integration is performed through Claude Code hook configuration, using PreToolUse hooks for policy enforcement and PostToolUse hooks for receipt signing. The setup runs locally and stores generated receipts in a project directory.
Typical use cases include regulated or high-assurance development workflows, compliance-focused projects, security-sensitive automation, and environments that require independently verifiable records of AI-assisted actions. The skill is particularly relevant for teams working in finance, healthcare, regulated research, or other domains where policy enforcement and audit evidence are required. It can also benefit developers, security engineers, compliance teams, and organizations seeking stronger governance over AI-driven tool execution.
It uses a PreToolUse hook to evaluate each tool invocation against a Cedar policy before execution. If the policy returns a deny decision, the tool call is blocked.
Receipts contain decision-related information including tool execution details, hashes of inputs and outputs, policy-related data, and timestamps. Receipts are signed and hash-chained.
Yes. The skill supports offline verification using the provided verification tooling, allowing third parties to validate receipts independently.
You need to install the plugin, configure Claude Code hooks in the project settings, create a Cedar policy file, and run the local receipt-signing server.
Policy evaluation depends on a Cedar policy configuration. The documented setup includes creating and referencing a policy file that governs tool execution decisions.
Quick Setup:
.claude/skills/Repository
wshobson/agents