Protect Mcp
Enforce Cedar policies and signed receipts on agent tool calls with Claude Code hooks and an MCP gateway before risky automation hits real systems.
Overview
ScopeBlind Protect is a MCP server for the Ship phase that applies Cedar policies and signed receipts to AI agent tool calls via hooks and an MCP gateway.
What is this MCP server?
- npm protect-mcp v0.5.4 with stdio MCP transport
- Signed receipts plus Cedar policies for AI agent tool calls
- Claude Code hooks integration and MCP gateway pattern from scopeblind-gateway repo
- Pairs with Passport (identity) and Verify (offline receipt validation)
- npm package protect-mcp version 0.5.4
- Repository: github.com/scopeblind/scopeblind-gateway
What problem does it solve?
Unbounded MCP tool access lets agents run destructive or non-compliant actions with no machine-verifiable record of what was allowed or denied.
Who is it for?
Solo builders shipping Claude Code workflows with multiple MCP tools who need policy enforcement and receipts, not honor-system prompts alone.
Skip if: Static sites or read-only research agents with no sensitive tools and no compliance requirement for signed invocation logs.
What do I get? / Deliverables
After install, tool calls can emit signed receipts and respect Cedar policies so you can audit and gate agent behavior before and during production use.
- Policy-gated tool call paths with signed receipts for allowed and denied actions
- Gateway configuration suitable for pre-production security review
- Evidence bundles verifiable offline with ScopeBlind Verify MCP
Recommended MCP Servers
Journey fit
Policy enforcement and signed receipts are ship-phase controls—you add them when tool access must be bounded, auditable, and defensible pre-production. Security subphase covers gates on what agents may invoke, aligned with review and launch-prep hardening rather than initial feature coding.
How it compares
Policy and receipt gateway MCP, not a vulnerability scanner or generic code-review skill.
Common Questions / FAQ
Who is com.scopeblind/protect-mcp for?
Indie operators and small teams running agentic coding stacks who need Cedar policies and signed receipts on MCP and hook-mediated tool calls.
When should I use com.scopeblind/protect-mcp?
Use it in ship and security prep when agents can call tools that change infra, secrets, payments, or customer data and you need deny rules plus evidence.
How do I add com.scopeblind/protect-mcp to my agent?
Install the protect-mcp npm package, register stdio MCP in Claude Code, configure hooks and gateway per scopeblind-gateway docs, and connect Passport identity where required.