What it does

Claude for Legal is a collection of plugins and reference agents for in-house legal teams, covering commercial, privacy, employment, litigation, regulatory, IP, corporate, and AI governance work. Each plugin runs as a Claude Cowork/Claude Code plugin or through the Managed Agents API, with the same system prompts and skills either way. The agents do named jobs: Vendor Agreement Reviewer, DSAR Responder, Termination Reviewer, Docket Watcher, and roughly three dozen others.

The interesting bit

The guardrails are the product. Every output is explicitly framed as a draft for attorney review, with source attribution on citations, conservative defaults on privilege, surfaced jurisdiction assumptions, and hard gates before anything can be filed or sent. Anthropic is unusually direct that these tools “do not replace” lawyers and “do not represent Anthropic’s legal positions.” In a field full of AI tools implying they can substitute for counsel, this reads as either refreshing honesty or careful risk management—possibly both.

Key highlights

• Two deployment modes from one source: desktop plugin via Claude Cowork/Code, or API-driven managed agents behind your own orchestration
• MCP connectors for both general productivity (Slack, Google Drive, Box) and legal-specific systems (Ironclad, DocuSign, iManage, Everlaw, CourtListener)
• Each practice area starts with a cold-start interview that learns your playbook, plus a shared CLAUDE.md profile that all skills in that area reference
• Scheduled “eyes-on-the-feed” agents for renewals, dockets, regulatory feeds, and diligence tracking
• Named agents with single-command invocation: /commercial-legal:review, /privacy-legal:dsar-response, etc.

Caveats

• Requires Claude Cowork access or self-hosted API deployment; not a standalone tool
• The README is comprehensive but the actual implementation depth per agent is unclear from documentation alone
• Some agent descriptions (e.g., litigation demand drafting with FRE 408 awareness) suggest sophistication that would need verification in practice

Verdict

Worth evaluating if you run an in-house legal team already using Claude’s ecosystem and want structured, opinionated starting points rather than building prompts from scratch. Skip if you’re looking for a self-contained legal AI product or don’t have the Claude infrastructure in place.