We built the first DMCA takedown MCP: counsel-backed enforcement your AI agent can call

The request that used to start with a Google search now starts with a sentence spoken to an AI assistant: “Someone copied my course and is selling it — can you get it taken down?” The person no longer opens ten tabs to learn what a DMCA notice is. They ask their agent to handle it.

For a takedown practice, that changes the question. For twenty years the goal was to be found — to rank for “remove stolen content.” The new goal is to be callable: to be a tool the agent can actually use when the person asks. So we built one.

What we shipped

Lawyerd now runs an MCP server — lawyerd-takedown — that exposes our enforcement work as tools any AI assistant can call. MCP (the Model Context Protocol, the open standard behind this new wave of agent tooling) is the same plumbing that connects assistants to email, calendars and code. We used it to connect an assistant to a takedown. Three tools, in plain terms:

• resolve_host — give it an infringing URL and it tells you who really hosts it and where a notice would go: the CDN or proxy in front (Cloudflare and the rest), the platform, or the direct host and its abuse address. For hosts we have revealed before, it returns the real one immediately. It is read-only and files nothing — a lookup, not a filing.
• assess_takedown — it lays out how a takedown for that URL would proceed: where the notice goes, what proof and attestation you would need to supply, the step-by-step process, and the legal caveats. It does not judge whether your claim is good — that is a human’s job — it plans the route.
• start_takedown — when you want to actually proceed, it hands back a secure onboarding link. The rights holder opens it once, in a browser, to identify themselves, sign a standing authorization, and show proof of ownership.

That last step is the important one, and it is deliberate.

The line we do not cross

Earlier this year the cautionary story made the rounds: an autonomous agent filed a copyright takedown against a project, and the supposed rights holder had no idea it had happened. That is the failure mode of pointing an agent at enforcement without a human in the loop — and it is a real legal hazard, because a takedown notice is sworn under penalty of perjury. Knowingly false notices carry liability under 17 U.S.C. §512(f). Software should not be making that oath on anyone’s behalf.

So our design draws a hard line. The agent can do everything except sign. It looks up the host, assesses the route, and initiates the matter — but the sworn statement is made once by the rights holder, a human, at onboarding. From then on the agent can track the matter without touching the oath again. Filings aimed at lawful criticism, genuine reviews, journalism or regulator notices are declined at intake; we challenge what is fake, infringing or impersonating, not what is merely unflattering. The convenience is the agent’s; the responsibility stays with the person and, above them, with counsel.

This is the same principle that governs the rest of the practice: automation is the engine, not the product. The MCP is a faster door into counsel-led enforcement — not a self-serve machine that files in your name while you sleep.

Why it is more than a lookup

A tool that only names a host would be a toy. Behind these three tools sits the enforcement pipeline we already run: revealing the real host even when it hides behind Cloudflare, filing the formal notice to that host, following up on a schedule when they go quiet, and — the part most services skip — confirming the content is actually gone before a matter is closed. Success means removed, not “notice sent.” The agent is a new way to reach that pipeline; the pipeline is the thing that has been taking content down for real clients.

The host-reveal piece is the one competitors don’t have as an agent tool. Most infringers sit behind a proxy precisely so the rights holder can’t find whom to notify. Answering “who really hosts this?” in a single call — and then acting on it — is the capability the whole thing is built around. (We wrote up the manual version of that work in how to find the real host behind Cloudflare.)

The claim

To our knowledge, as of July 2026, this is the first DMCA/takedown practice made callable by AI agents through MCP. There are excellent automated takedown platforms, and there are law firms beginning to expose services to agents — but we have found no one exposing takedown enforcement as an agent tool, and certainly no one doing it while keeping the rights holder’s signature and a human intake screen intact. If someone shows us otherwise, we will happily correct the record; the honest hedge matters more to us than the flag.

We think being callable is the new being indexed. The agentic web is arriving on real infrastructure — the standard for tools (MCP), and the payment rails for what agents consume (the direction Cloudflare and others are building toward). When a creator’s assistant can find the host, plan the notice and hand its owner a single link to authorize it, the distance between “someone stole my work” and “it’s being taken down” collapses to one conversation. That is the distance we wanted to close.

Try it

The resolve_host lookup is read-only and free of consequences — point an MCP-compatible client at the server and ask it who hosts a URL:

claude mcp add lawyerd-takedown https://mcp.lawyerd.net/mcp

Ask your assistant to resolve the host for a URL, or to assess a takedown for something that copied your work. When you’re ready to actually file, it will hand you the onboarding link — and from there, the practice takes it the rest of the way.

If you are a brand or a creator with content being copied, impersonated or cloned, you can also skip the agent and request a confidential audit: we’ll surface live abuses against you, signed by counsel, without obligation.

---

Informational, not legal advice. A takedown notice is a sworn legal document; grounds and forms differ by jurisdiction (US, EU/EEA, UK), and no outcome is guaranteed. Lawyerd does not remove lawful, genuine reviews or criticism — we challenge content that is fake, infringing, or impersonating.