aegis-zk

AEGIS Protocol

Community aegis-zk
Updated

Zero-knowledge attestations of agent skills and MCP servers.

AEGIS Protocol

On-chain zero-knowledge skill attestation for AI agents on Base

What is AEGIS?

AEGIS is a protocol for verifiable AI agent skill attestation. Auditors evaluate AI agent skills, generate zero-knowledge proofs of their assessment, and submit the results on-chain to the AEGIS Registry on Base. Anyone can query, verify, and dispute these attestations — creating a trustless reputation layer for AI agents.

How it works:

  1. A developer registers a skill definition (what the agent claims to do)
  2. An auditor evaluates the skill and generates a ZK proof of their assessment
  3. The attestation (proof + result) is submitted on-chain to the AEGIS Registry
  4. Anyone can verify the proof on-chain or dispute a fraudulent attestation

How to Use

AEGIS is a trust verification layer for AI agent skills — it does not execute skills. Use it to check whether a skill has been audited before you run it.

import { AegisClient } from '@aegisaudit/sdk';

const aegis = new AegisClient({ chainId: 84532 });

// 1. Discover registered skills
const skills = await aegis.listAllSkills();

// 2. Check attestations for a skill
const attestations = await aegis.getAttestations(skills[0].skillHash);

// 3. Verify the ZK proof on-chain
const trusted = await aegis.verify(skills[0].skillHash, 0);

// 4. If trusted → execute the skill using the publisher's own SDK/API

The typical integration flow:

  1. Query AEGIS — is this skill registered? Has it been audited?
  2. Verify the proof — is the audit cryptographically valid?
  3. Check the stake — how much ETH did the auditor risk on this assessment?
  4. Execute the skill — get the code from the skill publisher (not from AEGIS) and run it

See the SDK README for a full integration guide with audit levels.

Architecture

aegis/
├── packages/
│   ├── sdk/            # @aegisaudit/sdk — TypeScript client library
│   ├── mcp-server/     # @aegisaudit/mcp-server — MCP tools for AI agents
│   ├── contracts/      # Solidity smart contracts (Foundry)
│   ├── circuits/       # Noir ZK circuits (Barretenberg)
│   └── cli/            # Command-line interface
├── apps/
│   └── web/            # Frontend — React + Vite + Three.js
└── scripts/            # Deployment & seed scripts

Quick Start

git clone https://github.com/aegis-zk/aegisprotocol.git
cd aegis
pnpm install
pnpm build

Requires Node.js 20+ and pnpm 9+.

Packages

Package Description npm
@aegisaudit/sdk TypeScript SDK for querying and interacting with the AEGIS Registry npm
@aegisaudit/mcp-server MCP server exposing AEGIS as tools for Claude, Cursor, and other AI agents npm

Deployed Contracts

Contract Network Address
AegisRegistry Base Sepolia 0x851CfbB116aBdd50Ab899c35680eBd8273dD6Bba

Tech Stack

  • Language — TypeScript, Solidity, Noir
  • Blockchain — Base L2 (Ethereum rollup)
  • ZK Proofs — Noir circuits compiled with Barretenberg (BB.js)
  • Smart Contracts — Foundry (forge)
  • Client — viem
  • AI Integration — Model Context Protocol (MCP)
  • Frontend — React 19, Vite, Three.js, wagmi
  • Build — Turborepo, pnpm workspaces, tsup

Links

License

MIT

MCP Server · Populars

MCP Server · New

    civyk-official

    WinWright

    Playwright-style MCP server for Windows desktop, system, and browser automation. 110 tools for WPF, WinForms, Win32, Chrome/Edge via Model Context Protocol.

    Community civyk-official
    mavdol

    Capsule

    A secure, durable runtime for AI agents. Run untrusted code in isolated WebAssembly sandboxes.

    Community mavdol
    easyshell-ai

    EasyShell

    Lightweight server management & intelligent ops platform with Docker one-click deployment, batch script execution, web terminal, and AI-powered operations.

    Community easyshell-ai
    AVIDS2

    Memorix

    Cross-Agent Memory Bridge Persistent memory for AI coding agents across 10 IDEs (Cursor, Windsurf, Claude Code, Codex, Copilot, Kiro, Antigravity, OpenCode, Trae, Gemini CLI) via MCP. Team collaboration, auto-cleanup, mini-skills, workspace sync. Never re-explain your project again.

    Community AVIDS2
    zw008

    VMware AIops

    VMware vCenter/ESXi AI-powered monitoring and operations. Two skills: vmware-monitor (read-only, safe) and vmware-aiops (full operations) | Claude Code Skill

    Community zw008