Verifi

MCP serverAI & models

Lets your agent send its work to a real human for review and get feedback or decisions back.

Available today. Use it from your connected AI after setup.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use Verifi to verify claim

About this server

Get a real human to verify, decide, or improve an AI agent's work. Pay per review via x402 on Base.

Install Verifi

The server’s own address, for the clients that take one directly. Or connect ahel once and every client you use reads it from one address, with the account kept on ahel rather than in each client’s config.

  • Claude Code

    claude mcp add --transport http --scope user verifi 'https://verifi.cloud/mcp'

    Run it once in your project, then open /mcp to approve any sign-in the server asks for.

  • Claude Desktop

    https://verifi.cloud/mcp

    Add a custom connector in Settings, paste this address, and approve the sign-in.

  • Cursor

    cursor://anysphere.cursor-deeplink/mcp/install?name=verifi&config=eyJ1cmwiOiJodHRwczovL3ZlcmlmaS5jbG91ZC9tY3AifQ==

    Open the link and Cursor adds the server at that address.

  • ChatGPT

    https://verifi.cloud/mcp

    In Settings, enable Developer mode, create an MCP app, and paste this address. Your plan and workspace must allow custom apps.

  • Codex

    codex mcp add verifi --url 'https://verifi.cloud/mcp'

    Run it once, then sign in with codex mcp login verifi if the server asks for an account.

From the project's README

As published by jyharju-code/verifi in README.md.

Verified human loops for AI agents. An agent POSTs an intent and a claim; a real human reviews it and answers true, false, or a refined free-text correction. Payment is a native x402 micropayment (USDC or EURC on Base) behind an HTTP 402 paywall: no accounts, no API keys.

Live at verifi.cloud · API docs · llms.txt · MCP endpoint at https://verifi.cloud/mcp

Official MCP Registry name: cloud.verifi/human-verification. Publication and domain ownership details are documented in docs/MCP_REGISTRY.md.

Why

Every agent framework hits the same wall: the agent cannot trust its own output. Verifi gives the agent a durable human workflow, paid the way agents pay (x402), and designed to remain reliable when a good answer takes several minutes.

How it works

agent ──POST /verify──▶ verify-api ──▶ core engine ──▶ Telegram bot ──▶ human
  ▲                        │x402           │ route,        │ buttons:      │
  │                        │verify+settle  │ audit         │ ✅ ❌ ✏️       │
  └──────── answer ◀───────┴───────────────┴───────────────┴───────────────┘
  • Three gates, priced in euros: 0.10 EUR admits the request to the human queue. When the human answers within 60 minutes of admission, the result unlocks for 2.90 EUR; a later answer, within 24 hours, unlocks for 1.45 EUR. The price is fixed by when the human answered, never by when the agent unlocks. Pay in EURC, or in USDC converted at the ECB reference rate and rounded up to the cent. The admission asset binds the chain, and every price is in the first 402.
  • No free human work: every chain pays both gates. Human time is the scarce resource and wallet addresses are free to mint, so a free tier would let anyone flood the queue unpaid. Free without a human: the docs, MCP discovery, and the 402 response that lets an agent test the connection and read the price.
  • Failure credit: a chain whose admission was actually paid and nobody answers within 24 hours grants one free admission for the next chain, in the same asset. The unlock is not included, and credits cannot be farmed for free work.
  • Reliable delivery: every call returns 202 with a durable verify_id. Agents poll for as long as the human needs, up to 24 hours, or use the optional ready or failed callback.
  • MCP: agents can call verify_claim, get_verify, unlock_verify, and verifi_info directly.

Repository layout

core/          Python core engine
  api/         internal FastAPI + admin dashboard + public /contact
  bot/         Telegram bot for human responders (associates)
  mcp/         MCP server (Streamable HTTP)
  routing/     associate selection
  payments/    earnings, settlements, payouts
  db/          PostgreSQL schema + migrations
instances/
  verify-api/  public Verify API (Node.js + @x402/express)
  ask-this-finn/  second instance (placeholder)
deploy/        docker-compose, nginx, facilitator config, static site
scripts/       ops wrapper (lock + audit + turn-taking)

One VPS runs everything: PostgreSQL (+pgvector), the core engine, the bot, the Verify API, the x402 facilitator, the MCP server, and nginx.

Running it yourself

cp deploy/env.server.example .env   # fill in tokens and keys
docker compose -f deploy/docker-compose.yml --env-file .env up -d postgres
docker compose -f deploy/docker-compose.yml --env-file .env up -d --build \
  core-api verify-api mcp
docker compose -f deploy/docker-compose.yml --env-file .env --profile bot up -d bot
docker compose -f deploy/docker-compose.yml --env-file .env --profile payments up -d facilitator
docker compose -f deploy/docker-compose.yml --env-file .env --profile edge up -d nginx

You need: a Telegram bot token, a receiving wallet address (X402_PAY_TO), and a small gas wallet for the facilitator (FACILITATOR_PRIVATE_KEY, a few euros of ETH on Base). See docs/WALLETS.md for the money architecture and docs/DECISIONS.md for design decisions.

API in 20 seconds

curl -X POST https://verifi.cloud/verify \
  -H "Content-Type: application/json" \
  -d '{"intent": "send_email",
       "claim": "The user opted in via double opt-in on 2026-07-15.",
       "agent_id": "0xYourWalletAddress"}'

Every admitted response starts with 202 and status: "processing". Poll until ready or failed; a ready result is retrieved through the separate unlock endpoint. The canonical repository reference is docs/API.md, rendered at verifi.cloud/docs.

License

MIT

Tools it offers (4)

What this server listed when ahel dialed its public endpoint in Sep 2026, with no key and no account of yours. The names are the server’s own.

  • verify_claim
  • get_verify
  • unlock_verify
  • verifi_info

Signals

Last commit
Oct 2026
Advanced
Delivery
human-verification MCP server → your ahel connector (mcp.ahel.ai) → your AI.
Item type
mcp-server
Key
cloud-verifi-human-verification
Source
github.com/jyharju-code/verifi
Hosted endpoint
https://verifi.cloud/mcp