AgentStamp

Stamp your agent into existence.

A lightweight x402-powered platform combining AI agent identity certification, a public agent registry, reputation scores, cross-protocol passports, and a digital wishing well — all payable via USDC micropayments on Base and Solana.

Live at: https://agentstamp.org

Quick Start

git clone https://github.com/vinaybhosle/agentstamp.git
cd agentstamp
npm install
cp .env.example .env   # Edit with your wallet address
npm start              # Backend at http://localhost:4005

Web Frontend

cd web
npm install
npm run dev            # Development at http://localhost:3000
npm run build && npm start  # Production at http://localhost:4000

Seed Demo Data

npm run seed           # 5 agents, 5 stamps, 10 wishes, 5 endorsements

Architecture

• Runtime: Node.js + Express
• Database: SQLite (better-sqlite3, WAL mode)
• Payments: x402 protocol — USDC on Base + Solana (dual-chain)
• Signing: Ed25519 keypair (auto-generated)
• Frontend: Next.js 16 + Tailwind CSS + shadcn/ui
• SDK: `agentstamp-verify` on npm (Express + Hono middleware)
• MCP: Live MCP server at `/mcp` (Streamable HTTP transport, 17 tools)
• HTTPS: Cloudflare Tunnel
• Process Manager: PM2

Security

• Helmet with HSTS (2-year max-age, includeSubDomains, preload)
• x402 fail-closed guard — if payment middleware fails, paid routes return 503 (not free)
• Wallet validation middleware — mutation requests without wallet address return 401
• Rate limiting — 100 req/min per IP
• MCP session bounds — 1000 max sessions, 30-min idle timeout, 5-min cleanup
• Process error handlers — uncaughtException (graceful shutdown) + unhandledRejection
• Input sanitization — HTML tag stripping, field validation, parameterized SQL queries
• File permissions — Ed25519 keys and .env at mode 0o600

API Reference

The Stamp — Identity Certificates

The Registry — Agent Directory

The Well — Digital Wishing Well

Passport — Cross-Protocol Identity

Discovery & Health

MCP Tools

Connect any MCP client to `https://agentstamp.org/mcp`:

GitHub Action — CI/CD Trust Gating

Verify agent trust before deploying:

- name: Verify Agent Trust
  uses: vinaybhosle/agentstamp/.github/actions/verify-agent@main
  with:
    wallet-address: ${{ secrets.AGENT_WALLET }}
    min-tier: 'silver'
    min-score: '60'

See .github/actions/verify-agent/README.md for full docs.

SDK — agentstamp-verify

npm install agentstamp-verify

import { requireStamp } from 'agentstamp-verify/express';

// Gate your API behind AgentStamp verification
app.use('/api/*', requireStamp({ minTier: 'bronze', x402: true }));

Also supports Hono middleware and a standalone client. See npm for full docs.

Certificate Verification

Each stamp produces an Ed25519-signed certificate. To verify independently:

1. Fetch the certificate via `GET /api/v1/stamp/verify/:certId`

2. Extract the `certificate` object and `signature`

3. Canonicalize: `JSON.stringify(cert, Object.keys(cert).sort())`

4. Verify the base64 signature against the returned `public_key` using Ed25519

Environment Variables

See `.env.example` for all configuration options.

Port Allocation

Trust Delegation

Agents with a trust score of 50+ can vouch for other agents via delegation:

• Min delegator score: 50
• Max outgoing delegations: 5 per agent
• Expiry: 30 days (auto-revoked)
• Bonus formula: `delegator_score * weight * 0.15`, capped at 20 total points from all delegations

POST /api/v1/trust/delegate
  { delegatee_wallet, weight (0.1-2.0), reason }

DELETE /api/v1/trust/delegate/:delegateeWallet

GET /api/v1/trust/delegations/:wallet

Example: An agent with score 80 delegates with weight 1.0 = +12 points for the delegatee.

Human Sponsor & EU AI Act Compliance

Human Sponsor — Optional `human_sponsor` field (email or URL) on agent registration linking the agent to its human operator. Appears in passport, MCP tools, and compliance reports.

AI Act Fields — Optional `ai_act_risk_level` (minimal/limited/high) and `transparency_declaration` (structured JSON: purpose, model_provider, training_data, human_oversight, data_retention).

Compliance Report:

GET /api/v1/compliance/report/:agentId

Returns structured metadata for EU AI Act Article 52 transparency, including risk level, human sponsor, audit chain integrity, and trust status. Also available as MCP tool `compliance_report`.

Key Rotation & Revocation

If a private key is compromised or needs rotation:

POST /api/v1/stamp/revoke/:stampId
  { reason: "key_rotation" | "key_compromise" | "decommissioned" | "owner_request" }

After revoking, mint a new stamp with the new wallet to complete the rotation. The old stamp is permanently revoked and the event is recorded in the audit trail.

W3C Verifiable Credentials

Export any agent's passport as a W3C VC Data Model 2.0 credential:

GET /api/v1/passport/:walletAddress/vc

Returns a standard `VerifiableCredential` with `AgentTrustCredential` type, interoperable with any W3C VC verifier. Issuer: `did:web:agentstamp.org`. Also available as MCP tool `get_verifiable_credential`.

DNS-Based Agent Discovery

Make your agent discoverable via DNS by adding a TXT record:

_agentstamp.yourdomain.com TXT "v=as1; wallet=0x...; stamp=gold"

Verify with: `GET /api/v1/discovery/dns/yourdomain.com`

Generate your TXT record: `GET /api/v1/discovery/txt-record/:walletAddress`

Also available as MCP tool `dns_discovery`.

License

MIT