Anthropic launched Claude for Legal on Tuesday, a package of 12 practice-area plugins and more than 20 MCP connectors that wire its chatbot directly into the software law firms already run on. The release builds on a narrower legal plugin the company shipped in February, the one that briefly torched legal tech stocks before anyone had really used it.
Anthropic is now competing with its own customers
Plugins cover the obvious practice areas: Commercial, Corporate, Employment, Privacy, Litigation, IP, AI Governance, and several more. Each one starts with a setup interview that asks how a team actually works (playbooks, escalation chains, house style) and writes a profile Claude reads from later. The GitHub repo lists the full set, including managed-agent "cookbooks" that can be deployed through the Claude API.
The MCP connectors are where the strategic awkwardness lives. Harvey, the legal AI startup that raised $200 million at an $11 billion valuation in March, is now a connector inside Claude. So is Legora, which closed a $600 million Series D last month and ran a Jude Law ad campaign while doing it. Thomson Reuters' CoCounsel Legal connects too, and the company has rebuilt it on the Claude Agent SDK. The foundation model and the applications built on top of it are, increasingly, the same product sold by different parties.
"The legal sector is facing mounting pressure to adopt AI," an Anthropic spokesperson said, which is the kind of line a vendor delivers when it would prefer customers feel late.
Anthropic also claims legal professionals became the most-engaged users of its Cowork desktop agent after February's release, with usage over three times any other knowledge-work function. That figure comes from the company itself, with no methodology attached, and "knowledge-work function" is a category Anthropic gets to define. Take it as directional.
The court hallucination problem
Whether shipping more capable legal AI helps or hurts a profession that has spent the past two years getting publicly humiliated by it is the harder question. 404 Media catalogued eighteen lawyers caught filing AI-generated documents stuffed with fabricated case citations. California fined an attorney last year for an appeal riddled with fake quotes drafted in ChatGPT. Federal judges have been caught doing it too.
Anthropic's pitch is that practice-area plugins, research-tool connectors, and citation tagging make this less likely. The GitHub README notes that citations pulled through a connected research tool get tagged with their source, while citations from "model knowledge alone" are flagged with a [verify] note. The architecture admits something most legal AI pitches won't: the model still hallucinates, and the connectors exist to constrain it.
Useful concession. Whether it works in practice is another matter.
Money and timing
The launch lands a week after Anthropic shipped a similar package for financial services and ahead of a reported IPO window. The 2026 revenue run-rate sits above $30 billion, up from $9 billion last year, per figures it shared. Anthropic is variously reported at a $380 billion valuation and entertaining offers above $900 billion, depending on which round you look at.
A free webinar walking through Claude for Legal runs Friday. The plugins and connectors are available today to paying Claude customers.
