A law office run through Claude

lawyer.ai is AI-native practice management for law offices: matters, statutory deadlines, hearings, time and billing, drafting and cited legal research, driven through the assistant the firm already uses. The assistant proposes; a lawyer confirms. In production with its first firm.

Book a 30-minute demo How we build MCP servers

A deadline missed by one day is not a small mistake. It is the whole matter.

97
MCP tools drive the whole practice
38
guarded writes: preview, then confirm
1M+
court decisions in the research corpus
5 min
case-law research that used to take days

The problem lawyer.ai solves

A law office runs on deadlines that forfeit rights when missed. In most firms they are still computed by hand from the statute into a calendar, checked by whoever is not in a hearing that day. The office is never one person: leave, sick days and handovers decide whether an appeal deadline is watched or orphaned, and the partner finds out afterwards.

Research is its own tax. Case law, the official gazette and consolidated statute texts live on separate public portals with search that punishes a single typo, so a question that should take minutes takes a day. And the general-purpose AI tools that promise to fix this write confident fiction: a fabricated citation in a filing is not an inconvenience, it is a disciplinary matter.

lawyer.ai puts the whole practice behind one assistant, with the guardrails a law office actually needs: deadlines are computed, never guessed; citations are verified against the source, never invented; and nothing is written to the record until a lawyer has seen a preview and explicitly confirmed it.

One judgment, end to end

The best way to understand lawyer.ai is to follow a single served judgment through the office: read, computed, confirmed, researched, and filed, with the assistant driving and a lawyer deciding.

Book a 30-minute demo How we build MCP servers

Works with Claude, on the firm's own subdomain.

01

Deadlines are computed, never guessed

Enter the service date and the statutory basis; the engine does the rest. The term starts the day after service, skipped days are named, and the court recess is flagged for the lawyer's judgment, never silently applied. A reminder ladder fires at 7, 3 and 1 days.

02

The assistant proposes. A lawyer confirms.

All 38 write tools are two-step: a preview of exactly what would change plus a confirmation token first, the commit only on explicit sign-off. Raw IDs never leave the server, and every call is logged against the real signed-in user.

03

Research that cannot invent a citation

Over a million court decisions, the complete official gazette and consolidated statutes, connected by a citation graph. Every answer carries its source, and a draft check blocks any reference it cannot verify before it reaches a filing.

04

Filed on your letterhead, after your approval

Drafts stay drafts until a lawyer approves them; an approved document is immutable and stamps who and when. The finished work renders on the firm's own letterhead, down to branded workbooks and charts.

05

Research that took days takes minutes

Measured with the pilot firm and running in production since July 2026: what used to be a day or two of portal work comes back in minutes, every citation carrying its source.

Ask for it in plain language. Every corner of the practice answers.

97 tools cover the office end to end. Reads answer instantly; anything that changes the record is previewed and confirmed. A few of the things the firm's team actually asks:

Today

What's on my desk today?

daily_dashboard read-only

Deadlines in statutory-risk order, today's hearings and who attends, unrouted mail.

Deadlines

Any limitation dates at risk this quarter?

limitations_watch read-only

The statute-of-limitations dates across every open matter, each with its owner.

Hearings

Brief me for Monday's hearing. Vera is on leave.

hearing_brief approval-gated

A brief assembled from the whole file for the substitute, approved before use.

Billing

Log 1.5 hours of drafting on the lease matter and issue the invoice.

log_time · generate_invoice two-step

Tariff-based cost statement, numbered invoice with VAT, both held for confirmation.

Drafting

Draft the response to the opponent's filing from our template.

generate_draft approval-gated

A draft on the firm's letterhead; it stays a draft until a lawyer approves it.

Documents

File this scanned judgment to the matter.

read_document · file_to_matter two-step

Read with OCR, classified as a court decision, the appeal deadline proposed from service.

Research

Has the civil-procedure statute changed since we opened this case?

statute_diff · watchlist read-only

An article-level diff between editions, and an email the next time it moves.

Coverage

Who covers V. Hartley's matters in August?

office_coverage read-only

Absences, per-matter handover notes and reassigned hearings. The office is never one person.

In production, not in a deck

lawyer.ai runs a real practice today, and it is not a first attempt.

  • A European law firm runs its matters, deadlines, hearings and billing on lawyer.ai
  • MCP connector live in production since July 2026, on the firm's own subdomain
  • Case-law research measured with the pilot firm: 1 to 2 days of portal work, now minutes
  • Built by Proceptio, the studio operating five production MCP servers with 200+ governed tools
See the case studies

Frequently asked questions

What is lawyer.ai?
lawyer.ai is an AI-native practice management platform for law offices, built by Proceptio. It runs matters, clients, statutory deadlines, hearings, tasks, time, cost statements, invoicing, drafting and legal research, and it exposes the whole practice to Claude through an embedded MCP connector with 97 tools, so the firm's team can drive it conversationally.
Can the assistant really change data in our practice?
Only through a two-step gate. Every one of the 38 write tools first returns a preview of exactly what would change, plus a confirmation token computed over that exact payload; nothing is written. The change commits only when the lawyer explicitly confirms and the second call carries the token. Read tools are read-only, and every call, read or write, is logged against the real signed-in user.
How do you stop invented citations?
Three ways. Research answers always carry the source they came from. A citation guard verifies every cited decision and statute against the corpus rather than trusting the model. And a whole-draft check reviews a document's citations before filing, returning "not found in the corpus" for anything it cannot verify instead of letting an invented reference through.
What does the research corpus cover?
Over a million court decisions reaching back to 1988, at parity with the public portal it draws from, plus the complete official gazette and the consolidated texts of the active statutes, connected by a citation graph of over half a million edges. Search is full-text and typo-tolerant, and a regulation watchlist emails the firm when a followed law changes.
Is our firm's data used to train AI?
No. lawyer.ai does not ship a model and holds no model API key; the assistant is the firm's own Claude, connected through the firm's own subdomain. Practice data stays in the firm's tenant, research answers are grounded in the public-law corpus, and nothing about the connector sends your matters anywhere else.
Which jurisdictions does it support?
lawyer.ai is built end to end for one European civil-law jurisdiction: its statutory deadline engine, court catalog, tariff-based billing and full research corpus are specific to it by design, because that depth is exactly what generic tools lack. The architecture underneath is jurisdiction-portable. If you want it for yours, talk to us.
How does a firm connect Claude to it?
Each firm gets its own subdomain and a connector URL to add in claude.ai or Claude Desktop. Lawyers sign in with their existing work identity, and the same per-user permissions that govern the browser govern the assistant. Setup is minutes, not a project; the connector is built on MCP, the open standard Anthropic launched for exactly this. Read about how we build MCP servers.