Descrybe

What it is

Descrybe is a US legal research engine — case-law search, a citator, statutes and regulations, quote and citation checking — that you can buy as its own web app or as a $25-a-month data layer plugged into Claude, ChatGPT, Microsoft 365 Copilot or Perplexity. It answers one question: can a solo, a small firm or an in-house team get grounded case-law research inside the AI assistant they already pay for, without a Westlaw or Lexis seat? For US primary law, yes, at the lowest published price in the category.

The company is two co-founders in Newton, Massachusetts — Kara Peterson (CEO) and Richard DiBona (CTO) — and describes itself as self-funded. It began as a free tool for plain-English case summaries and is now a paid product. The coverage page lists 10,987,388 opinions from 492 courts and tribunals as of 11 September 2026, plus statutes, regulations, constitutions, session laws and court rules; the company’s headline figure is 300 million structured primary-law records.

2026 is when distribution became the product. Descrybe was one of the launch collaborators for Anthropic’s Claude for Legal on 12 May 2026, shipping the “Descrybe Legal Engine for Claude” connector. It reported more than 700 new paid subscribers in the six weeks that followed. Then came the ChatGPT App Directory (2 July), an open-source Python SDK called Open Connector for firms building their own research tools (late July), availability in Microsoft 365 Copilot (September), and a ChatGPT Enterprise plugin announced on 17 September.

Why it shows up in Legal Ops stacks

• The Legal Engine is the cheapest paid research connector you can put behind an assistant. $25 per user per month buys 13 research tools — primary-law search, the Cytator citator, citation resolution, quote checks — usable from every supported assistant on one subscription, with no per-search charges. Midpage, the closest peer, publishes $100.
• It catches the failure lawyers fear most. Quote verification and citation resolution are exposed as separate tools, so an agent drafting inside Claude or ChatGPT can check that a quoted line exists in the opinion it cites before the text reaches a brief.
• One subscription across assistants. A firm split between ChatGPT users and Copilot users does not buy two research products. That is rarer than it sounds: most legal AI vendors pick one host.
• Open Connector lets you own the workflow. The SDK packages Descrybe’s research steps as reusable Python, so a firm’s own internal tool can call the same engine instead of routing everything through a chat window.
• Academic distribution builds familiarity early. NELLCO approved it as an e-resource in March 2026, reaching close to 150 law libraries, and it reports use in 350+ university curricula. Associates arriving from school may already know it.

Pricing reality

Legal Engine is $25 per user per month; the full Descrybe Platform is $50. Both renew monthly. The Platform includes the Legal Engine plus the Descrybe workspace, DescrybeLM Deep Research, saved history and Advanced Brief Checking, capped at 150 full research runs a month. The Legal Engine lists no usage cap. There is no published free trial; a LAUNCH50 code halves the first month, and payments are non-refundable. Organizations get central billing at a flat price per named user, quoted on request; universities buy through NELLCO.

The real cost is the assistant underneath. The Legal Engine does nothing without a Claude, ChatGPT, Copilot or Perplexity subscription, billed separately. A solo on Claude Pro plus the Legal Engine pays roughly $45 a month. Five litigators on the full Platform cost $3,000 a year — half of five Midpage seats, and under an eighth of five Westlaw Precision with CoCounsel seats at roughly $428 each.

Best for

Solo attorneys, small litigation firms, legal-aid organizations and in-house teams who already work inside Claude, ChatGPT or Copilot and need US case law with verifiable citations and quote checks, at a price a single practitioner can expense without a procurement cycle.

Alternatives and when to pick them

• Thomson Reuters CoCounsel with Westlaw — the market-share leader. Pick it when you need secondary sources, KeyCite by name, or a vendor your malpractice carrier and procurement team already know.
• LexisNexis Protégé — the other incumbent. Pick it when a firm-wide Lexis agreement exists and Shepard’s is your standard.
• Harvey — the fastest-growing entrant in legal AI. Pick it for firm-wide transactional and advisory work across a large team, not for a solo’s research budget.
• Midpage — the closest like-for-like. Pick it at $100 when you need agentic PACER docket access, a SOC 2 Type II report, or versioned point-in-time statutes.
• CourtListener — pick it when the budget is zero. Its free MCP server grounds an assistant in case law, but it has no citator, no quote-check tool and no statutes.
• Paxton — pick it when you want one app for research, drafting and medical chronologies rather than a data layer inside your assistant; it lists at $499 a month.

If a court rule, your carrier or opposing counsel expects KeyCite or Shepard’s, Descrybe does not replace the incumbent. Run it on the side and let the seat count settle the renewal.

Watch-outs

• No SOC 2 report yet. Descrybe’s own security page says it is in SOC 2 readiness work and claims neither SOC 2 nor ISO 27001; it relies on its providers’ certifications, with OpenAI’s API as the primary model provider. Guard: keep client-identifying facts out of research queries until the report exists, and ask for the SOC 2 timeline in writing before any firm-wide rollout.
• The benchmark is self-reported. The March 2026 claim that DescrybeLM answered all 200 multistate bar questions while ChatGPT, Claude and Gemini missed 13 to 23 comes from Descrybe’s own study, with no independent replication found. Guard: run your last ten research memos through it and check every authority against your current citator before you rely on the Cytator.
• Coverage data has visible errors. The coverage page lists some court end dates in 2035 and 2041, and several Texas Court of Appeals districts start only in late 2025. Guard: look up the specific courts you litigate in on the coverage page during your first paid month, and confirm a recent opinion from each resolves.
• Two founders, no outside capital. Self-funding means no investor pressure to sell, but also a thin bench against Thomson Reuters and RELX. Guard: stay month-to-month, keep research memos with their full citations in your own DMS, and do not let an incumbent contract lapse on Descrybe alone.
• Your research surface depends on the host. Connector behaviour changes when Anthropic, OpenAI or Microsoft change their platforms. Guard: keep the Descrybe web app as the fallback path if you pay for the Platform, and re-test a known query after major host updates.

For the surrounding category, see best legal AI research tools, best MCP tools for legal ops and grounding vs hallucination in legal AI.