The DMV for AI Agents: A Community Namespace with Real Ambition

The internet never solved identity for humans. For AI agents, the window to get it right is closing now.

Agent Community launched the Department of Machine Verification (DMV) on June 26, a free pre-registration system for .agent domain names. The project, described on its site, lets builders register a name for their AI agent and receive a content-addressed certificate ID. The registration flow is simple: run bunx dmv-agent register in a terminal, provide a name and operator email, and get back a verifiable identity. The community has 29,000 members and 7,000 companies backing it, according to the Launly listing.

This is not a blockchain play. It is not a Web3 land grab. It is a bet that the naming layer of the agent internet should be community-governed, open, and applied for through ICANN before a single large company captures the namespace.

The timing matters. 2025 and 2026 have seen explosive growth in autonomous AI agents. OpenAI, Anthropic, Google, and a dozen startups ship agents that browse the web, execute code, and transact on behalf of users. Every one of those agents needs an identity that other agents and services can verify. Without a standard, the default is fragmentation: each platform issues its own IDs, every integration requires bespoke trust, and the big platforms win by default.

DMV’s approach is pragmatic. It uses a DNS TXT record for discovery: _agent.name.agent. 300 IN TXT "v=aid1;uri=https://your-endpoint;p=mcp". The Agent Identity and Discovery (AID) spec, hosted at aid.agentcommunity.org, defines how an agent publishes its endpoint and protocol support. The DMV itself handles registration and verification, with MCP tools built in for dmv_doctor, register_agent, and verify_certificate.

The project is applying to ICANN for the .agent Top-Level Domain. That is a slow, expensive, and uncertain process. ICANN’s new gTLD program requires a US$185,000 evaluation fee per application, plus ongoing registry costs in the hundreds of thousands annually. Agent Community’s 29,000 members and 7,000 companies provide community support but not necessarily the capital. The Launly analysis rates feasibility as Medium, with “high complexity, substantial legal/compliance costs, and regulatory risks.”

Pre-registration is smart. It builds a user base and a set of claimed names before ICANN decides. If the .agent TLD is approved, pre-registrants have a strong claim to their names. If it is denied, the community still has the DMV certificate system and the AID spec, both of which work independently of the TLD.

The competitive landscape is messy. ENS and Unstoppable Domains offer blockchain-based name services. Worldcoin offers identity verification through biometrics. Decentralized identity (DID) standards exist but lack adoption. None of these are built specifically for AI agents. DMV’s advantage is focus: it is designed for the agent use case, with MCP tool support, a CLI that works in Claude Code, and a DNS-based discovery mechanism that any agent can implement.

The risk is that a platform captures the namespace before the community does. OpenAI could issue agent IDs through its own API. Anthropic could build identity into Claude’s agent framework. Google already has the infrastructure. If any of these companies ship a widely adopted agent identity system, the community-governed alternative becomes a niche.

DMV’s answer is openness. The AID spec is a standard, not a product. The DMV registration is free. The identity cards are shareable as PNGs or links. The community owns the governance. That is a strong pitch to builders who do not want their agents locked into a single platform’s identity system.

The real test is adoption. 29,000 members is a start, but the agent economy needs millions of identities to function. DMV needs agent frameworks to integrate the AID spec. It needs agent marketplaces to verify certificates. It needs the ICANN application to succeed or a fallback that works without it.

The project’s name is deliberately playful. DMV as the Department of Machine Verification echoes the bureaucratic tedium of human identity. But the problem it addresses is serious. Every agent that calls an API, every agent that negotiates with another agent, every agent that signs a transaction needs a verifiable identity. The internet’s naming layer for humans is broken. The agent internet does not have to repeat the mistake.

The question is whether the community moves faster than the platforms.