MCPS -- The Secure MCP

Zero Trust for AI Agents

Interactive playground showing MCPS, AgentSign, AgentPass, and rag-secure working together. Register an agent, sign messages, query a verified knowledge base, and see trust-gated access in action.

Trust Authority L3 PASS AgentSign Verify / Sign / Gate AGENT Cursor AGENT Gemini AGENT Bot Bad PROTECTED MCP Server PROTECTED API Gateway BLOCKED SIGNED SIGNED
Interactive Demo

Full Agent Security Flow

Click each step in order. Watch the audit trail build.

1
AgentSign

Register Agent

Issue a cryptographic identity. Agent starts at L0 (untrusted).

2
MCPS

Sign an MCP Message

Every tool call gets an ECDSA P-256 signature with nonce and timestamp.

3
MCPS

Verify the Signature

Receiving server verifies the signature, nonce, timestamp, and agent identity.

4
rag-secure

Query RAG with Integrity

Retrieve documents with cryptographic verification and injection scanning.

5
AgentPass

Try Sensitive Operation (Blocked)

Attempt a write operation that requires L2 trust. Agent is L0 -- will be blocked.

6
AgentPass

Upgrade Trust Level

After verification, the agent earns higher trust. Upgrade one level at a time.

L0 L1 L2 L3 L4
7
AgentPass

Retry Sensitive Operation

After upgrading to L2+, the same operation succeeds.

8
Audit Trail

View Hash-Chained Audit Log

Every action is logged with cryptographic hashes. Tamper-evident by design.

The Stack

Every layer. Signed. Verified. Audited.

LayerProductWhat it doesStandard
IdentityAgentSignECDSA P-256 agent identity certificatesPatent pending
SigningMCPSPer-message signing, nonce, replay protectionIETF Submission
TrustAgentPassTrust levels L0-L4, graduated access controlPatent pending
RAG Integrityrag-secureDocument signing, verification, injection scanningnpm
TransportATTPSigned HTTP for agent-to-agent communicationIETF Submission
ScanningCybersecifyOWASP Top 10 scan, DAST, deep source analysisOWASP