ooligo

Claude for Legal

legal-ai-assistant agent-platform · legal-workflow-automation · mcp-connectors
AI-NATIVE MCP API
Legal Ops
8.8 /10

What it is

Claude for Legal is Anthropic’s legal vertical, launched 12 May 2026: 12 practice-area plugins, 20-plus MCP connectors into the systems legal teams already run, and scheduled background agents that fire on a cadence instead of waiting for a prompt.

The thing to understand before you evaluate it is that there is no SKU. Claude for Legal is not a separate product with its own login, contract, or price line. It is an Apache 2.0 plugin suite published at github.com/anthropics/claude-for-legal that installs into a Claude seat you already pay for — Claude Cowork, Claude Code, or the Managed Agents API for headless runs. The model vendor moved into the workflow layer without asking anyone to buy anything new, and that is what makes it a different kind of competitor to Harvey and CoCounsel than another legal AI startup would be.

What you actually get

The 12 plugins are cut by practice area: commercial, corporate, employment, privacy, product, regulatory, AI governance, IP, and litigation, plus law-student, legal-clinic, and a legal-builder-hub for teams writing their own.

The connectors are the load-bearing part. Document management (iManage, NetDocuments), contract systems (Ironclad, DocuSign, Definely), eDiscovery (Everlaw, Relativity, Consilio), deal rooms (Box, Datasite), research (CourtListener, Trellis, Midpage, Descrybe), and — the pairing that surprised the market — rival legal AI platforms, with connectors for Harvey and Thomson Reuters CoCounsel Legal reaching Westlaw Deep Research.

Scheduled agents are where it stops being a chat product. renewal-watcher reads your CLM and surfaces auto-renew windows before they close; docket-watcher polls dockets; reg-change-monitor tracks regulatory movement against your policies; ip-renewal-watcher and dataroom-watcher do the same for filings and diligence.

The scoped use case that justifies it

A four-person in-house commercial team drowning in renewal deadlines. Install with /plugin install commercial-legal@claude-for-legal, run the required /commercial-legal:cold-start-interview once to capture playbook positions, and point renewal-watcher at your Ironclad instance. You get a weekly list of contracts inside their notice window with the deviations from playbook already flagged — work that otherwise lives in a spreadsheet nobody updates. Setup is an afternoon, not a procurement cycle.

Pricing reality

The plugins cost $0 and always will — Apache 2.0. What you pay for is the seat that hosts them, and there is no free Cowork tier:

  • Pro — $20/mo, $17/mo billed annually
  • Max — from $100/mo (5× Pro usage) to $200/mo (20×)
  • Team — $25/seat/mo, $20 annually; premium seats at $125/$100 carry 5× the usage; adds SSO and admin controls
  • Enterprise — $20/seat for access with model usage billed at API rates; adds SCIM, audit logs, and custom data retention

The bill that catches people is not Anthropic’s. Connectors to iManage, NetDocuments, Ironclad, Everlaw, Relativity, and CoCounsel run on your subscriptions and your API keys. A 10-lawyer team on annual Team seats is about $2,400/year; the DMS and CLM contracts those connectors depend on are an order of magnitude above that, and they are the prerequisite, not the upsell. If you license none of them, Claude for Legal reaches nothing but the documents you upload by hand — which is a real use case, but a much smaller one than the launch coverage suggested.

Best for

In-house legal teams — GC, legal ops manager, commercial counsel — that already license a DMS or CLM and want agentic workflows over them without signing for a second AI platform. Also the right pick for firms and clinics that need to change the prompt rather than fill in a settings page: Apache 2.0 means you fork a practice-area plugin, and the Legal Clinic plugin was built inside ABA Formal Op. 512 guardrails.

Versus the alternatives

Harvey is the segment leader and the safer answer for large firms — pick it when you want a vendor accountable for output quality, firm-wide onboarding, and a support contract, and when its LexisNexis alliance — which puts Shepard’s citation validation inside Harvey — matters to your litigators. The head-to-head against CoCounsel walks that decision. Thomson Reuters CoCounsel is the incumbent to pick when primary law is the job and you are already paying for Westlaw — CoCounsel’s authority coverage is the thing Claude for Legal connects to rather than replaces. Legora is the fastest-growing entrant and the pick when you want a collaborative drafting surface your associates will actually open, rather than an assistant they have to remember to invoke.

Take Claude for Legal over all three when your constraint is budget and control rather than coverage: you have engineering or legal-ops capacity in-house, you want to own the prompt, and your work is workflow orchestration across systems you already run — not primary legal research.

Watch-outs

  • Context does not carry across matters. Each plugin depends on a CLAUDE.md file plus a cold-start interview per plugin, and that setup is per-user, per-install. Ten lawyers means ten interviews and ten drifting playbook files. Guard: commit one CLAUDE.md to a shared repo or drive and make it the single source, then treat playbook edits as reviewed changes rather than personal notes.
  • Citations are unverified unless you wire a research connector. Without CourtListener, Trellis, Descrybe, or CoCounsel attached, the output flags cites [verify] and means it. Guard: attach a research connector before anyone relies on a citation, and treat any [verify] marker as a hard block on the document leaving the building.
  • Everything it produces is a draft for attorney review — Anthropic says so in the repo, and the plugins gate before anything is filed or sent. Guard: name the reviewing attorney per workflow in your intake process, the same way you would for a paralegal’s first draft.
  • Privileged material routed through a general-purpose assistant is a governance question, not a feature question. Guard: run it on Enterprise, where custom data retention and audit logs exist, and get the retention setting and the connector scopes signed off by whoever owns your outside-counsel guidelines before the first matter goes through it.