NEW: Listen to our podcast -- MCP is Dead. Long Live MCPS.Listen on Spotify
IETF Internet-Draft: draft-sharif-mcps-secure-mcp · Standards Track
AGENTCursor AGENTGemini AGENTBot Bad AgentSignVerify / Sign / Gate PROTECTEDMCP Server PROTECTEDAPI Gateway Trust Authority SIGNED SIGNED L3 PASS BLOCKED
Zero Trust for Agentic AI

Your agent needs an identity

Zero trust identity for autonomous AI agents. Every action signed. Every tool gated. Every agent verified. No identity, no trust.

Agent security is not optional. OWASP Agentic Top 10 and the EU AI Act now require it.

0
Runtime Dependencies
3
API Calls to Ship
12
Frameworks Audited
<10KB
Package Size
agentsign -- live demo
Built for teams running autonomous AI agents
Claude MCP
Gemini ADK
LangChain
CrewAI
AutoGen
OpenHands
Composio

None of these frameworks have agent identity built in. Your agents are running unsigned, unverified, and unaccountable. See the audit.

Standards & Recognition

Built on open standards

AgentSign is grounded in IETF specifications, OWASP guidance, and published intellectual property. Not a wrapper -- the standard itself.

OWASP
MCP Security Cheat Sheet

Contributor to Section 7: Message Integrity & Replay Protection. The official OWASP guidance for securing Model Context Protocol deployments.

cheatsheetseries.owasp.org →
IETF Internet-Draft
draft-sharif-mcps-secure-mcp

MCPS: Cryptographic security layer for Model Context Protocol. Message signing, trust verification, and replay protection for MCP tool servers.

datatracker.ietf.org →
IETF Internet-Draft
draft-sharif-agent-payment-trust

AgentPass: Challenge-response identity and trust framework for autonomous AI agent payments. Cryptographic proof before any financial transaction.

datatracker.ietf.org →
IETF Internet-Draft
ATTP -- Agent Trust Transport Protocol

The secure transport layer for AI agents. Mandatory cryptographic signing, identity, and audit on every agent-to-server API call. No insecure mode.

datatracker.ietf.org →
Intellectual Property
UK Patents Filed

Patent applications filed at the UK Intellectual Property Office covering agent identity, MCP security, agent transport, ATTP, and payment authorisation.

UKIPO -- Patent Applications Filed
npm Package
agentsign@1.1.2

Zero-dependency Node.js SDK. Agent identity, passport signing, MCP Trust Gate, and verification in one package. Works with any framework.

npmjs.com/package/agentsign →
The Gap

600,000+ GitHub stars.
Zero agent identity.

We audited 12 popular agent frameworks. None have cryptographic agent identity, execution signing, or trust scoring.

Framework Stars Identity Signing Trust Score MCP Gate Revocation
AutoGPT182K
LangChain / LangGraph100K+
MCP Ecosystem80.7K
OpenHands64K
Microsoft AutoGen50.4K~
CrewAI45.6K
HuggingFace smolagents25.5K
OpenAI Agents SDK19.4K
Google ADK / Vertex15.6K~~
NeMo Guardrails5.7K
Amazon BedrockManaged~~
Devin / Cognition AIClosed
AgentSignOSS

= native support    ~ = partial (cloud-locked IAM)    = not available    Source: GitHub, March 2026.

600,000+ agents running without identity = 600,000+ potential attack vectors

The EU AI Act (August 2026) mandates traceability for high-risk AI systems. If your agents can't prove who they are, you're not compliant.

Product

Identity in three API calls

Register, onboard your agent, verify anywhere. No OAuth, no dashboard required. Pure API.

Without this, your agent is an anonymous process with production access. That violates OWASP ASI03.

Cryptographic Identity

Every agent gets a unique identity backed by HMAC-SHA256 signatures. Register, onboard, and go. The agent carries its signed passport everywhere.

  • One POST to register, one POST to onboard
  • HMAC-SHA256 signed passport returned instantly
  • 6-stage pipeline: INTAKE to ACTIVE
  • Trust score computed from pipeline stage
  • Instant revocation -- kill switch
Try It Now
cURL
# 1. Create account curl -X POST https://agentsign.dev/api/auth/register \ -H "Content-Type: application/json" \ -d '{"email":"dev@company.com"}' # Returns: { api_key: "as_live_..." } # 2. Onboard your agent curl -X POST https://agentsign.dev/api/agents/onboard \ -H "Authorization: Bearer YOUR_KEY" \ -d '{"name":"my-agent","permissions":["read"]}' # Returns: { agent_id, passport: { ..., signature } }

Agent Passport

A self-contained, signed JSON document the agent carries everywhere. Any system can verify it offline -- no server needed. Like SSL certificates, but for AI agents.

  • Self-contained: ID, name, trust score, signature
  • HMAC-SHA256 signed -- tamper-proof
  • Offline verification -- no server roundtrip
  • Pipeline stage embedded
  • Revocation flag -- instant kill switch
Try It Now
JSON
{ "agent_id": "agent_a7f3b2e1...", "name": "my-agent", "permissions": ["read", "write"], "stage": "ACTIVE", "trust_score": 100, "issued_at": "2026-03-12T...", "issuer": "agentsign-platform", "version": "1.0", "signature": "f514d0b870..." } // Any system can verify this offline. // No server. No API call. Just crypto.

MCP Trust Gate

The identity layer between agents and MCP tool servers. Before any agent calls a database, file system, or API via MCP, the Trust Gate checks identity, stage, and trust score.

  • Agent presents ID to MCP server
  • Trust score threshold enforcement
  • Permission-based tool access
  • ALLOW / DENY on every tool call
  • Public endpoint -- no auth needed
Try It Now
cURL
# MCP server calls this before granting access curl -X POST https://agentsign.dev/api/mcp/verify \ -H "Content-Type: application/json" \ -d '{"agent_id":"agent_xxx","tool":"query"}' # Returns: { "decision": "ALLOW", "agent_id": "agent_xxx", "trust_score": 100, "stage": "ACTIVE" }

Passport Verification

Any service can verify an agent's passport in one API call. Check if the signature is valid, the agent hasn't been revoked, and the passport hasn't been tampered with.

  • One POST with the passport JSON
  • Validates HMAC-SHA256 signature
  • Detects tampering instantly
  • Public endpoint -- no auth needed
  • Returns full passport metadata
Try It Now
cURL
# Verify any passport curl -X POST https://agentsign.dev/api/verify \ -H "Content-Type: application/json" \ -d '{"passport": { ... }}' # Valid passport: { "valid": true, "passport": { ... } } # Tampered passport: { "valid": false, "passport": { ... } }
How It Works

Three steps to zero trust

Sign up, onboard your agent, verify everything. That's it.

1

Sign Up

Create a free account via the console or API. Get an API key instantly. No credit card required.

  • One POST or browser signup
  • API key returned immediately
  • Free tier: 5 agents included
  • No OAuth, no waiting
2

Onboard

Register your agent with one API call. It gets a cryptographic identity and signed passport.

  • POST /api/agents/onboard
  • Name, permissions, stage
  • Signed passport returned instantly
  • Works with any framework
3

Verify

Verify any agent's identity before granting access. Public endpoint, no auth needed.

  • POST /api/verify (signature check)
  • POST /api/mcp/verify (trust gate)
  • Works offline with passport
  • ALLOW or DENY in milliseconds
Ecosystem

Built on MCPS

Teams are already integrating MCPS into their security stack.

🛡
AgentShield
MCPS INTEGRATED

AI agent observability and governance platform. Real-time risk analysis, human-in-the-loop approvals, and cost attribution -- secured with MCPS cryptographic identity verification.

MCPS L0-L4 Trust LangChain CrewAI MCP
useagentshield.com →
+
Your Product Here

Integrate MCPS into your agent platform, MCP gateway, or security tool. Get listed in our ecosystem.

Become a Partner →
Pricing

Start free, scale when ready

No credit card required. Upgrade when you need more agents.

Free
$0
forever
  • 5 agents
  • 1,000 verifications/mo
  • 500 MCP gates/mo
  • HMAC-SHA256 signing
  • API + Console access
Get Started
Startup
$149
per month
  • 200 agents
  • 100,000 verifications/mo
  • 50,000 MCP gates/mo
  • Everything in Developer
  • SLA + dedicated support
Get Started

Need unlimited agents? Contact us for Enterprise pricing.

The cost of not signing your agents?

Data breaches, regulatory fines under the EU AI Act, and rogue agents with production access. The free tier covers 5 agents -- there is no reason not to start today.

Deployment

Your infrastructure, your keys

AgentSign runs wherever you need it. Start with our cloud to test, deploy on-prem for production. Same API, same SDK.

Cloud (agentsign.dev)

Get started in 30 seconds. We host the server. Ideal for testing, small teams, and prototyping.

  • Instant signup, no infrastructure
  • Free tier: 5 agents forever
  • Same API as on-prem
Start Free
Enterprise
🏢

On-Prem (Self-Hosted)

Deploy in your VPC. Your infrastructure, your signing keys, your data. Zero calls to us. Full compliance.

  • Docker / Kubernetes / bare metal
  • Your keys, your audit boundary
  • OWASP aligned / IETF standard
Contact Sales

Both options use the same SDK (npm install agentsign). Switch from cloud to on-prem with one config change.

Case Study

The agent identity crisis

We audited the MCP ecosystem. 518 servers scanned. Zero agent identity.

41%
Zero authentication
1,422
Unprotected MCP tools
0%
Agent identity in any framework

"AI agents inherit extensive permissions and cannot reliably distinguish legitimate instructions from attacker-injected content, enabling zero-click exfiltration, session hijacking, and credential vault takeover."

-- OWASP LLM01:2025 / TLDR InfoSec, March 2026

"83% of organizations plan to deploy agentic AI, yet only 29% feel prepared to secure it. Every framework authenticates the user. None authenticate the agent."

-- OWASP Agentic Top 10 (ASI03), December 2025

Real-World MCP Security Incidents

These are documented, published CVEs and breaches -- not hypotheticals:

CVE-2025-6514 (CVSS 9.6)
mcp-remote RCE -- 437K downloads affected
Smithery.ai Breach
3,243 MCP servers exposed, API keys compromised
Asana Cross-Tenant Leak
1,000 customers exposed for 34 days via MCP
postmark-mcp Backdoor
First malicious MCP server -- BCC'd all emails to attacker

The root cause in every case: no agent identity, no message signing, no trust verification. MCP has no security layer. These attacks will keep happening until the protocol has one.

AgentSign adds the missing identity layer. No valid passport = no access. Revoked = instant kill switch.

Full Compliance & Audit Report

Give your agents an identity

Three API calls. Five minutes. Zero trust from day one.

Every unsigned agent is a liability.

41% of MCP servers have zero auth. 6 CVEs in 9 months. Tool poisoning succeeds 72.8% of the time. The time to act is now.

Create Free Account Read Quickstart

No credit card required. 5 agents free forever. See compliance requirements.