Register an agent in two minutes. Give a site an agent front in fifteen.

An agent signs every request with its own key, under the open Web Bot Auth standard, and carries a Verigrant agent ID that says who operates it and what it is there to do. When it acts for a person, it also carries a pass that shows a verified person sent it.

A website checks all of that on its own server, against keys Verigrant publishes. It answers with an agent front: a live, structured API of what the site offers and the actions it accepts, served as plain HTTP and as MCP. Structured data both ways, instead of scraping one way.

verigrant agent register --email ops@example.com --name "Example Agents" --accept-terms \
    --agent-name shopper --purpose assist --network 203.0.113.0/24 --rate 500 \
    --abuse-contact abuse@example.com

What you declare for each agent:

1. A name.
2. The purposes it acts for, from a closed list: assist, acting for one person at a time; search, building a search index; research, reading for analysis without training; training, collecting for model training.
3. The model or framework it runs on.
4. The networks it sends from, or Verigrant's egress.
5. How many requests a day it expects to make to each site.
6. A contact for abuse reports.

What you get: a Verigrant agent ID, a signed credential that lives at most 24 hours, bound to your agent's key. The kit renews it for you. Your declarations are a promise you make in the terms you accept, and sites can hold you to them.

Climb the levels when you need to. Sites choose the lowest level they accept.

1. Level A. A confirmed email and your agent's keys. Verigrant hosts your key directory for you.
2. Level B. Prove your domain with one DNS line or one file, and publish your key directory there. Rechecked daily.
3. Level C. Your business checked against the public company register, with a card on file. The card is checked, not charged.
4. Level D. A named officer passes Verigrant's identity check.

Other commands: keys list, keys add, keys rotate, keys revoke, domain claim, domain verify, status and credential. There is an operator console on the web for the same things, plus the abuse reports against your agents and your standing.

The two minutes, and the fifteen below, are timed in the release test, core-api/tests/agent_network_e2e.rs, which fails if an operator takes longer to reach level A or a site takes longer to reach a live front.

Connect makes the pairing one click. Verigrant is an OAuth 2.1 authorization server for remote MCP clients, following the MCP authorization specification. The person signs in, reads your limits in plain words, and approves. Your agent then reaches Verigrant's MCP server with a short lived token. No keys to paste.

What your agent can then do through Verigrant:

1. Get a pass for one site.
2. Propose a task plan for the person to approve, and get the payment approval for a step that buys.
3. Collect the person's traveller details, sealed to one business. You carry them. You cannot read them.
4. Drop a complete order and read the business's signed outcome.
5. Receive booking updates the person allowed.

You never hold a readable copy of the person. A mistake on your side is an embarrassment, not a breach.

It also warns you when something does not add up: a request from a network the agent did not declare, a connection fingerprint that changed without a key rotation, a rate above what the agent declared, or a purpose your terms do not allow.

Every agent request is logged with its signature. The log stays on your own storage unless you choose hosted reporting.

The gate applies the verdict in front of your pages.

1. Registered agents are pointed to your front with a machine readable answer.
2. Unannounced automation is told to register at verigrant.com/agents.
3. Suspicious visitors get a one time proof of work per session.
4. Rate limits apply by network, by fingerprint and by session, so a scraper spread over many addresses is still seen as one.
5. Hidden trap links mark whatever follows them as automated.

Where it runs:

1. A Rust core, and the gate as a sidecar for nginx, Caddy and Traefik.
2. A native TypeScript package, with middleware for Express, Next.js and Cloudflare Workers.
3. Python is planned.

1. The verify library, in Rust and TypeScript.
2. The gate.
3. The door that serves an agent front.
4. The generator and the verigrant CLI.
5. The agent kit, in Rust and TypeScript.
6. The agent ID and agent front formats.

Run by Verigrant, and not open source: issuing agent IDs and passes, people's sealed data, the order drop, Fill with Verigrant, private compute and the egress.

Verification never calls home. A site checks agent IDs, passes and signatures against public keys it holds. Verigrant does not learn which agent visited which site unless the site chooses to report.

git clone https://github.com/VERIGRANT_GITHUB_ORG/verigrant
cargo install verigrant-cli
npm install @verigrant/verify
npm install @verigrant/gate
npm install @verigrant/agent-kit

Every site that proves its domain can list its agent front: its name, categories, location, address, and the levels and purposes it accepts. Agents search the directory to find where to start.

Every verified operator is listed too, with its level and its standing. Abuse reports with evidence lower an operator's standing.

The search is GET https://verigrant.com/api/directory/search, with no credential.

1. Mass fake operators. Level A is worth little to sites that matter. Registration is rate limited. Level B needs a domain, level C a card and a real business. Sites choose their minimum.
2. A stolen agent key. Credentials live a day at most and are bound to the key. Revocation reaches sites through a public status list.
3. A replayed request. The signature covers the site, the path and a fresh value within sixty seconds. A request signed for one site fails at another.
4. Lying about purpose. We cannot see intent. We make lying provable and expensive: signed logs, reports with evidence, and revocation of the whole operator.
5. A malicious website. Fronts are served from the site's own domain, listed only after domain proof, and every text field is data, never instructions.
6. Tracking a person across sites. The agent ID names the operator's agent, not the person. Passes carry a different address for the person at each site.
7. Evading the gate. No wall is perfect. Fingerprint and session limits, proof of work and traps make it slow and costly. Announcing is always cheaper.

Registering an agent is free, at every level. A card check at level C charges nothing.

The open source is free under Apache 2.0.

Level one at a website is free for the site and for the agent.

Businesses pay for level two, sealed data, and level three, hosting. People never pay.