Open Source -- MIT Licensed

The HTTPS of the Agent Era

Cryptographic identity, message signing, and trust verification for the Model Context Protocol.

MCP is HTTP. MCPS is HTTPS.

Read the Spec View on GitHub
$ npm install mcp-secure Copied!
$ pip install mcp-secure Copied!
0
Dependencies
43
Tests Passing
8/10
OWASP MCP Risks Mitigated
2
SDKs (npm + PyPI)
L0-L4
Trust Levels
The Problem

MCP Has No Security Layer

No identity. No message signing. No tool integrity. No revocation. Real CVEs exist today.

CVE-2025-6514 -- CVSS 9.6

No Identity

Any agent can connect to any MCP server. No way to verify who is calling. No passports, no certificates, no trust.

OWASP MCP Top 10

No Message Signing

JSON-RPC messages travel unsigned. Man-in-the-middle attacks can modify tool calls, inject parameters, or replay requests.

82% Vulnerable

No Tool Integrity

Tool definitions can be tampered with after registration. Tool poisoning and rug pulls are trivial. 82% of MCP servers have path traversal vulnerabilities.

Enterprise Blocker

No Revocation

Compromised agents cannot be revoked. No mechanism to blacklist bad actors. No audit trail. Enterprise calls MCP a "non-starter."

The Solution

What MCPS Adds to MCP

A cryptographic security layer that sits on top of MCP. Like TLS for HTTP.

Agent Passports

ECDSA P-256 signed identity credentials. Every agent gets a cryptographic passport that proves who they are.

Message Signing

Every JSON-RPC message wrapped in a signed envelope. Nonce + timestamp prevents replay attacks.

Tool Integrity

Signed tool definitions prevent poisoning and rug pulls. Detect tampering before execution.

Real-Time Revocation

Revoke compromised agents instantly via Trust Authority. Blacklist bad actors across the network.

Trust Levels (L0-L4)

From unsigned (L0) to audited (L4). Enforce minimum trust levels per server. Progressive trust.

On-Premise

Run your own Trust Authority. All verification stays local. Nothing phones home. Full enterprise control.

Quick Start

5 Lines to Secure Your MCP Server

Drop-in SDK. Available for Node.js and Python.

mcp-secure
const mcps = require('mcp-secure');

// Generate keys + passport
const keys = mcps.generateKeyPair();
const passport = mcps.createPassport({
  name: 'my-agent',
  version: '1.0.0',
  publicKey: keys.publicKey,
});

// Trust Authority signs the passport
const signed = mcps.signPassport(passport, taPrivateKey);

// Sign any MCP message
const envelope = mcps.signMessage(mcpMessage, passport.passport_id, keys.privateKey);

// Verify on the receiving end
const result = mcps.verifyMessage(envelope, keys.publicKey);
// { valid: true }
from mcp_secure import generate_key_pair, create_passport, sign_passport
from mcp_secure import sign_message, verify_message

# Generate keys + passport
keys = generate_key_pair()
passport = create_passport(
    name="my-agent",
    version="1.0.0",
    public_key=keys["public_key"],
)

# Trust Authority signs the passport
signed = sign_passport(passport, ta_private_key)

# Sign any MCP message
envelope = sign_message(mcp_message, passport["passport_id"], keys["private_key"])

# Verify on the receiving end
result = verify_message(envelope, keys["public_key"])
# {"valid": True}
Trust Framework

Trust Levels

Progressive trust. Enforce minimum levels per server.

LevelNameMeaningRequirements
0UnsignedPlain MCP, no MCPSNone
1IdentifiedPassport presentedValid passport with public key
2VerifiedPassport verified + not revokedTrust Authority confirmation
3ScannedVerified + passed security scanOWASP agentic AI scan
4AuditedScanned + manual auditTrust Authority manual review
Security Coverage

OWASP MCP Top 10 Coverage

MCPS mitigates 8 of 10 OWASP MCP security risks.

MCP01
Token Mismanagement -- Passport-based identity
MCP03
Tool Poisoning -- Tool integrity signatures
MCP04
Supply Chain -- Signed tool definitions
MCP06
Intent Subversion -- Signed messages
MCP07
Insufficient Auth -- Passport verification
MCP08
Lack of Audit -- Signed audit trail
MCP09
Shadow Servers -- Only passported agents
MCP10
Context Injection -- Envelope isolation
Live Scan Results

We Scanned 39 Agent Frameworks

OWASP Agentic AI Top 10 security assessment. Point-in-time static analysis. March 2026.

FAIL 13 agents
WARN 17 agents
PASS 9 agents
AgentCategoryStarsVerdictRiskKey Findings
Open InterpreterCode Agent57K+FAIL80exec(), os.system(), child_process, no sandbox
OpenHandsCode Agent69K+FAIL70subprocess, exec, shell commands, file write
AutoGPTAutonomous182K+FAIL65exec(), os.system(), no sandbox
GooseCode Agent32K+FAIL65Shell exec core, pkg install, no sandbox
DevikaCode Agent19K+FAIL65exec(), subprocess, browser automation
Browser UseBrowser Agent80K+FAIL60Arbitrary browser control, JS injection
BabyAGIAutonomous20K+FAIL55exec(), eval(), supply chain
GPT EngineerCode Agent53K+FAIL55exec(), subprocess, supply chain
AgentGPTAutonomous32K+FAIL55eval(), no sandbox, supply chain
MetaGPTMulti-Agent48K+FAIL50subprocess shell=True, no sandbox
ChatDevMulti-Agent27K+FAIL50exec(), subprocess
SuperAGIAutonomous15K+FAIL50subprocess shell=True, excessive agency
CAMELMulti-Agent6K+FAIL40eval(), supply chain
n8nWorkflow178K+WARN26Code execution nodes, shell commands, spawn()
LangChainFramework100K+WARN26Supply chain, prompt injection vectors
SWE-agentCode Agent18K+WARN26Docker exec, subprocess, shell commands
CrewAIMulti-Agent45K+WARN23Supply chain, excessive permissions
Microsoft AutoGenMulti-Agent50K+WARN23Code execution patterns, supply chain
TaskWeaverCode Agent5K+WARN23Eval patterns, no sandbox
LangflowAgent Builder145K+WARN23Inherits LangChain exec patterns
AiderCode Agent41K+WARN23subprocess for git/tests, shell execution
DifyAgent Platform132K+WARN21Sandboxed code exec, tool calling
Crawl4AIWeb Agent61K+WARN21Playwright subprocess, arbitrary URL fetch
Semantic KernelFramework23K+WARN18Supply chain, output handling
SmolagentsFramework15K+WARN18Code execution, supply chain
ComposioAgent Tooling27K+WARN18Sandboxed exec, shell tools
LettaStateful Agent21K+WARN18Tool execution, subprocess
OpenAI Agents SDKFramework19K+WARN11Supply chain, excessive agency config
HaystackFramework18K+WARN11Supply chain, output handling
SwarmMulti-Agent18K+WARN11Excessive agency, supply chain
LlamaIndexFramework38K+PASS3Minor supply chain note
Google ADKFramework15K+PASS3Minor supply chain note
Amazon Bedrock AgentsCloud Agent--PASS3Minor supply chain note
Vercel AI SDKSDK12K+PASS0No issues detected
Anthropic Claude SDKSDK8K+PASS0No issues detected
PhidataFramework18K+PASS0No issues detected
InstructorSDK9K+PASS0No issues detected
MastraFramework10K+PASS0No issues detected
AgentSign SDKIdentity SDKOSSPASS0No issues detected

Point-in-time static analysis. Not a certification. Scanned March 2026 by AgentSign.

View Full Registry
Full Specification

2,603 Lines. 17 Sections. 3 Appendices.

JSON Schema. Example flows. Everything you need to implement MCPS in any language.

Read MCPS-SPEC.md View Repository
Enterprise

Run Your Own Trust Authority

All keys stay local. All verification stays internal. Nothing phones home.

terminal
$ docker run -p 8080:8080 agentsign/server

# Your own Trust Authority.
# Your own keys.
# Nothing phones home.

Powered by AgentSign -- Zero Trust Engine for AI Agents

Secure your agents today.

Zero dependencies. Node.js + Python. MIT licensed.

npm install mcp-secure pip install mcp-secure Star on GitHub