Trademark watching for the agent era
An MCP-native agent that watches USPTO, EUIPO, WIPO Madrid, the five major TLDs, and X — and drafts the response before the 30-day window closes.
The status quo
Registration is the easy part. The hard part is the next ten years.
Every week the USPTO publishes new applications. Every day someone registers a domain one Levenshtein-edit away from your client's brand. Every hour a squatter claims a handle on X. The Trademark Official Gazette goes out on Tuesdays. Madrid Monitor's weekly XML feed lands on Mondays.
The mark is registered. The defense is theoretical.
The threat surface
The USPTO opposition window is 30 days from publication. UDRP filing windows are tight enough that a missed week is a missed remedy. DMCA notices lose force the longer infringing content stays up.
Manual monitoring costs you a paralegal. PDF watch services cost $1,000 – $10,000 per mark per year and ship reports nobody opens until Tuesday. Neither approach watches X. Neither catches a typo domain registered Saturday morning.
The cost of inaction compounds. It always lands on counsel.
How it works
trademarc is an agent — not a chatbot wrapped around TESS. It exposes 12 typed tools to any MCP-aware client (Claude Code, Cursor, Windsurf) and watches your portfolio across every P1 registry continuously.
One line in your agent of choice. The MCP server runs locally or as a hosted Worker.
Verb-first typed tools: register, update, soft-delete. Hash-chained audit log from the first write.
USPTO & EUIPO daily, Madrid weekly, domain zones daily, X handles hourly. Per-source circuit breakers.
Findings land in Slack, Asana, and email. Cease & desist, DMCA, UDRP, TTAB opposition — statutorily compliant on arrival.
Built for the lethal trifecta
Three things never belong in the same context window: private portfolio data, untrusted ingest from the open web, and the ability to take external action. A prompt injection in a scraped offender page can hijack a naive agent.
trademarc's architecture answers this directly — a quarantined Gemini Flash instance ingests untrusted content and emits typed JSON only; a Zod strict parser between models fails closed; Claude Opus runs privileged actions against the validated payload.
Defense in depth
High-stakes actions pass through a 3-of-3 multi-model gate — Anthropic, OpenAI, Google. Disagreement escalates to a human. PII is redacted before any third-party call.
Every URL routes through an SSRF egress proxy with allowlist, IP-block, DNS-rebinding defense, and disabled HTTP redirects. Every action emits a hash-chained audit row. 713 tests. Two Cloudflare Workers in production today.
What's in the agent
Quarantined Gemini Flash for untrusted ingest; Claude Opus for privileged actions. Zod strict parser between them. Free-text never crosses the boundary.
Cease & desist, DMCA notice, UDRP complaint, TTAB opposition — formatted to the relevant statute or rule, with citations, evidence references, and a counsel-review-ready signature block.
USPTO TESS daily, EUIPO eSearch Plus daily, WIPO Madrid weekly, ICANN CZDS zones for .com/.net/.org/.io/.ai daily, X handles hourly. Per-source circuit breakers with documented fallbacks.
Append-only audit log with structured fields, hash-chained writes, tombstone attestation. Every tool call, every gate decision, every redaction logged. Exportable for client review and bar inquiry.
SSRF egress proxy with allowlist, IP-block, DNS-rebinding defense, and disabled HTTP redirects. Per-tenant column-level encryption via AWS KMS. Server-side RBAC re-read from live DB inside every serializable write transaction.
High-stakes actions require quorum from Anthropic, OpenAI, and Google. Disagreement escalates to a human. PII redacted before any third-party LLM call. DPAs required for autonomous mode.
trademarc is the agent layer for trademark practice — surveillance, evidence, drafting, audit. Open source, MCP-native, designed for the way you and your agents already work. P1 is drafting-only on purpose; the human stays in the loop for every send.
Install trademarc on GitHub