MCP server: One MCP server.

AuthiChain

@undone0603MCP server · External

Look up AuthiChain product records. x402 paid verification and on-chain anchoring in development.

Fraud & scam detectionv1.3.1TypeScript
Freeno charge per call

One address for any MCP client. No sign-in needed on this server.

What it does

5 read

Grouped by tool name. At most: only reads or looks up information.

What it costs

Free

Odel charges nothing per call. The publisher may charge for its own service.

Tools

5 · grouped by tool name · checked 11 Oct 2026

Read · 5Look things up

  • get_pricing
  • verify
  • query_provenance
  • verify_record
  • dpp_readiness_check
Try it in the Inspector
README5 min read

AuthiChain Unified

The canonical implementation of AuthiChain's physical-world trust layer.

AuthiChain turns a physical product or asset into a verifiable digital identity: a standards-aware identifier resolves to a signed claim, policy and provenance data are evaluated, and an independent verifier returns an auditable result.

This repository is the canonical build/deploy surface for:

  • AuthiChain — verification, attestations, product identity, API and billing
  • QRON — beautiful, programmable verification experiences and QR studio
  • GovChain — government/contractor trust workflows and seals
  • StrainChain — provenance and product verification for regulated physical goods
  • AgentZ / MCP — agentic orchestration and machine-facing access to the same trust primitives

Core thesis: one verification protocol, many commercial surfaces. The vertical products should not fork the trust model.

Architecture at a glance

text
                    PHYSICAL WORLD
                          │
                 product / asset / unit
                          │
                          ▼
                 GS1 Digital Link
              GTIN + serial / lot / qualifiers
                          │
                          ▼
                  ┌───────────────┐
                  │   RESOLUTION  │
                  │ identifier →  │
                  │ representation│
                  └───────┬───────┘
                          │
                          ▼
              ┌──────────────────────┐
              │ AUTHICHAIN VERIFY    │
              │ signed claims        │
              │ issuer / JWKS         │
              │ status / revocation   │
              │ provenance / policy   │
              │ scan-risk signals     │
              └──────────┬──────────┘
                         │
              ┌──────────┼──────────┐
              ▼          ▼          ▼
           Humans      Agents     Enterprise
           /verify     MCP/API    integrations
              │          │          │
              └──────────┼──────────┘
                         ▼
              QRON · GovChain · StrainChain
                         │
                         ▼
                billing / onboarding /
                audit / operational data

Trust boundary

Resolution is not verification. A resolver answers what identifier/resource is being addressed. AuthiChain verification evaluates whether a signed claim is valid, current, attributable and consistent with the evidence available to the verifier.

A positive status is therefore scoped to the evidence and policy being evaluated. For example, a registered product identifier does not by itself prove manufacturing quality, and a missing identifier is not by itself proof of counterfeiting.

Repository map

LayerLocationResponsibility
Web / customer surfacesclient/, workers/QRON and vertical user experiences at the public domains
Verification protocolprotocol/Open-source protocol/reference implementation
Agentic operationsagentz/Workflow orchestration, pilots, operational automation
Machine interfacemcp/MCP access to AuthiChain capabilities
Edge/APIworkers/*Cloudflare Workers, routing, verification and integrations
PersistenceSupabase/Postgres + D1Operational state, attestations, events and application data
Architecture docsdocs/ESTATE.md, docs/NETWORK.mdSystem inventory and deployment topology

Public surfaces

  • authichain.com — protocol, verification, certificates, API and billing
  • qron.space — QRON creation and programmable experiences
  • govchain.us — government/contractor trust workflows
  • strainchain.io — provenance and product verification workflows

The domains are commercial and presentation surfaces over the same underlying trust architecture. The canonical customer paths are /onboard, /dapp, and /verify.

Standards and verification flow

  1. Identify — a physical unit is represented with a stable identifier; where applicable, use GS1 Digital Link syntax such as /01/{gtin}/21/{serial}.
  2. Resolve — the identifier resolves to the appropriate representation or linkset.
  3. Attest — an issuer signs claims about the identified object.
  4. Verify — the verifier checks signatures, issuer keys, status/revocation and applicable policy.
  5. Interpret — the system reports the evidence-backed state and any risk signals without conflating registration with physical inspection.
  6. Act — humans, agents or enterprise systems consume the result through web, API or MCP interfaces.

Deployment model

Cloudflare-first. The production edge is Cloudflare Workers and the repository's worker deployment configuration. Legacy framework code may remain in the repository where it supports compatibility or migration, but it is not the architectural center of gravity.

  • Edge/runtime: Cloudflare Workers + Wrangler
  • Web: Vite + React
  • Data: Supabase/Postgres and Cloudflare D1
  • ORM: Drizzle
  • Package manager: pnpm
  • Agent runtime: Python / AgentZ
  • Protocol/API components: TypeScript, Hono and standard cryptographic libraries

See docs/NETWORK.md for deployment topology and docs/ESTATE.md for the system inventory.

Getting started

bash
cp .env.example .env
pnpm install
pnpm dev

For database-backed scripts, never put credentials in source code. Set DATABASE_URL in the local environment or the deployment secret store.

Example:

bash
export DATABASE_URL='postgresql://USER:PASSWORD@HOST:5432/DATABASE?sslmode=require'

AgentZ

bash
python -m agentz.cli list
python -m agentz.cli run authichain_pilot_deploy --mode dry-run

Pilot readiness

The canonical readiness path is:

install → typecheck → lint → tests → production build → deploy smoke test → real product scan

See:

  • docs/operations/PILOT-READY-BASELINE.md
  • docs/attestation/v0.1.md
  • docs/ESTATE.md
  • docs/NETWORK.md

Security

Deploy invariant

A push to main, including a docs-only commit, runs deploy-cloudflare.yml and deploy-edge-worker.yml. Do not add paths or paths-ignore.

Secret scan (gitleaks 8.28.0, --no-git) runs first and must fail closed. Deploy permissions stay contents: read. No token values in Git.

A green Wrangler deploy means that commit's bundle was published. It is not a SOC 2, FedRAMP, FDA, government, or cannabis COA certificate. A verifier pass is the checks this checkout ran.

authichain-consensus-engine is an identity collision. Confirm it in Cloudflare before any retirement. This file does not retire it.

Security is part of the protocol presentation, not an afterthought.

  • No credentials in Git. Runtime secrets belong in environment/deployment secret stores.
  • No live secrets in examples. Use placeholders such as sk_live_…, whsec_…, or postgresql://USER:PASSWORD@HOST/DB.
  • Rotate exposed credentials immediately. Removing a secret from the current tree does not invalidate a credential that may already exist in Git history, caches or logs.
  • Verify before release. Run secret scanning before merging and keep the existing security checks enabled.
  • Least privilege. Prefer narrowly scoped credentials and service bindings over broad account tokens.

If a credential has ever been committed, treat it as compromised until the provider confirms rotation/revocation.

Project principle

AuthiChain is not fundamentally a blockchain UI. The durable product is the verification layer: standardized identity → resolution → signed attestation → independent verification → evidence-backed decision.

Blockchain anchoring, NFTs, QR art, AI agents and vertical applications can extend that layer; they should not redefine its trust boundary.

License

See docs/project/LICENSE.md.

Similar modules

Browse Fraud & scam detection