DEV Community

Edison Flores
Edison Flores

Posted on

UTA — the universal trust layer for ALL AI agents (not just one)

UTA — the universal trust layer for ALL AI agents

Every AI agent tool has the same problem: how do you trust an MCP server before loading it?

Tool MCP support Trust verification
Cline ❌ None
Cursor ❌ None
Continue.dev ❌ None
Aider ❌ None
OpenHands ❌ None
LangChain ❌ None
CrewAI ❌ None

UTA fixes this for ALL of them. One package, universal compatibility.

What UTA provides

  1. 8 credential format support — ATC v3, JWT, W3C VC, A2A, EAT-AI, ZTA, MCP Card, X.509
  2. 12-stage fail-closed pipeline — UNKNOWN = DENY, ERROR = DENY, EXPIRED = DENY, REVOKED = DENY
  3. Pre-exec filter — blocks .env reads, rm -rf, shell spawns, credential exfiltration
  4. Merkle tree audit log — tamper-evident, signed
  5. Behavioral drift detection — detects when MCP server behavior changes over time
  6. Attack chain detection — catches multi-step attacks across tool calls
  7. Memory poisoning detection — detects injected instructions in tool responses
  8. Cross-agent trust — Agent A verifies Agent B before delegation

Install

# TypeScript
npm install @marketnow/trust-core @marketnow/trust-adapters @marketnow/trust-gateway

# Python
pip install marketnow-trust

# Rust
cargo add marketnow-trust

# Go
go get github.com/alicelabs-llc/universal-trust-adapter/uta-monorepo/packages/uta-go
Enter fullscreen mode Exit fullscreen mode

Live API (no auth, no signup)

GET https://www.marketnow.site/api/trust?action=formats
POST https://www.marketnow.site/api/trust?action=verify
POST https://www.marketnow.site/api/trust?action=translate
Enter fullscreen mode Exit fullscreen mode

Stats

  • 460+ tests passing
  • 36 test vectors
  • 6,744 verifications/sec
  • 9,248 MCP servers scanned (0.3% malicious)
  • 14 features shipped (v5.1-v6.0 complete)

Links

— Edison Flores, AliceLabs LLC
https://www.marketnow.site · https://status.marketnow.site

Top comments (0)