Tools
Claude for Legal Plugins Explained: The 12 Practice-Area Plugins and 20+ Connectors
Claude for Legal gives paid Claude users 12 open-source practice-area plugins, 150+ skills and 20+ connectors to iManage, NetDocuments and CoCounsel.
On this page
Claude for Legal is Anthropic's open-source set of 12 practice-area plugins for Claude, released on 12 May 2026 and free to paid Claude users. Each plugin bundles slash-command skills, scheduled agents and hard gates, and the 20+ MCP connectors reach iManage, NetDocuments, DocuSign, Relativity, Everlaw, CoCounsel, LexisNexis and Harvey. It runs in Cowork, Claude Code, Managed Agents and Microsoft 365, and every output is labelled a draft for attorney review.
The day a plugin moved the market
On 3 February 2026 Anthropic published a blog post announcing a legal plugin for Claude Cowork. By the close of trading Thomson Reuters was down 18% intraday, RELX 14% (its steepest single-day decline since 1988) and Wolters Kluwer 13% (Legal.io). Within three weeks Harvey and LexisNexis had both announced Anthropic integrations. The plugin itself was modest: contract review, NDA triage and compliance checklists inside a desktop agent. What moved the market was the demonstration that a foundation-model lab could ship legal workflows directly to lawyers at chatbot prices.
The 12 May 2026 release turned the demonstration into a product line: 12 plugins, 20+ connectors, an Apache 2.0 licence and a public GitHub repository (LawSites). Freshfields, which had signed a multi-year Anthropic deal on 23 April covering 5,700 staff in 33 offices, reported roughly 500% growth in Claude usage in six weeks, more than 20,000 people registered for Anthropic's April legal webinar, and Anthropic said legal professionals had become the most engaged Cowork users of any function (Artificial Lawyer). As of 18 September 2026 the repository lists 150+ skills, 13 scheduled agents and 20+ connectors and has about 9,500 stars.
What a plugin actually is
A Claude plugin is a folder. It contains skills (Markdown files with a short front-matter block and instructions, invoked as slash commands such as /commercial-legal:review), scheduled agents that run on a timer, hooks that fire before or after an action, and connector definitions that point at MCP servers. The skills and plugins entry covers the mechanics; the practical point is that the whole thing is readable text. A partner who can read a playbook can read, and edit, the skill that applies it.
Two conventions run through every plugin. First, each one begins with /<plugin>:cold-start-interview, a 10–20 minute structured conversation that writes a CLAUDE.md practice profile: your jurisdictions, your standard positions, your escalation rules. Every other skill reads that file, so a plugin without an interview produces generic output. Second, the repository's disclaimer is blunt: "Every output is a draft for attorney review — NOT legal advice", and "These plugins do not represent Anthropic's legal positions" (GitHub).
The twelve plugins, one line each
| Plugin | What it does | Notable control |
|---|---|---|
| commercial-legal | MSA, NDA and SaaS review to Word tracked changes; RED/YELLOW/GREEN triage; amendment history; 90-day renewal tracker | Agents renewal-watcher, deal-debrief and playbook-monitor |
| corporate-legal | Tabular diligence with "every cell cited"; disclosure schedules; closing checklists; board consents; entity compliance | Review gate before board consents execute; detects Luminance and Kira output for QA |
| employment-legal | Hiring and termination review; worker classification; FMLA/CFRA/PFL leave tracker; privileged investigation memos | Privilege framing on investigation memos |
| privacy-legal | DPA review; DSAR responses within statutory clocks; PIA/DPIA triage; policy monitor | Deadline-aware DSAR skill |
| product-legal | Launch review; marketing-claims substantiation; "is this a problem?" Slack triage; Linear/Jira/Asana | Triage routes rather than answers |
| regulatory-legal | Feed watcher; policy diff; gap tracker; NPRM comment deadlines | Redrafts are proposals, not edits |
| ai-governance-legal | EU AI Act per-system inventory; AI impact assessments; vendor-AI term review; policy starter | Built for the AI Act deployer duties |
| ip-legal | Trademark clearance; FTO triage; cease-and-desist; DMCA §512(g); OSS licence check; portfolio renewals | Send gate on takedowns |
| litigation-legal | Matter intake; demand letters; subpoena triage; legal holds; claim charts; chronology; deposition prep; privilege-log review; brief sections; docket-watcher via CourtListener, Trellis and Everlaw | FRE 408 gate on demand letters; issue/release gates on holds |
| legal-clinic | Built around ABA Formal Opinion 512; supervisor review queue; student ramp; intake; deadlines | Nothing leaves without supervisor sign-off |
| law-student | Socratic drilling, outlining, IRAC, bar prep | "Learning mode, not answer mode" |
| legal-builder-hub | Discover and install community skills | Injection scan, licence gate and allowlist before install |
A thirteenth, partner-maintained plugin, cocounsel-legal, gives Claude access to Westlaw Deep Research through the CoCounsel Legal MCP and needs a CoCounsel subscription. Five agents (diligence-grid, docket-watcher, launch-radar, reg-monitor, renewal-watcher) also ship as headless Managed Agents cookbooks that run in Anthropic's cloud on a schedule without a desktop.
Connectors: the document store finally talks back
The plugins are the visible part. The connectors are what make them useful in a firm, because they let Claude read from the systems where the documents already live rather than from files dragged into a chat. As of September 2026 the list includes iManage, NetDocuments, DocuSign, Ironclad, Relativity, Everlaw, Box, Datasite, Thomson Reuters CoCounsel, LexisNexis, Harvey, Midpage, Trellis and Definely, all speaking the Model Context Protocol.
The two DMS connectors set the governance pattern. iManage's MCP Server (14 May 2026) gives "governed, permission-bound, fully logged" access that "respects existing ethical walls" and works with Claude, ChatGPT, Copilot, Harvey and Legora alike (iManage). NetDocuments' ndMAX connector runs every call under the user's own credentials, applies matter restrictions automatically, keeps documents inside the platform and can be switched off with a single admin toggle (NetDocuments). That is the right shape: the agent inherits the lawyer's permissions, not a service account's. The DMS connection guide walks through the setup.
| Connector | Type | What it enables |
|---|---|---|
| iManage, NetDocuments, Box | Document stores | Search and read matter files under existing permissions and walls |
| CoCounsel (partner plugin), LexisNexis, Midpage, Trellis, CourtListener | Research | Citation checking; without one, citations are tagged [verify] |
| Relativity, Everlaw | E-discovery | Docket and evidence access for litigation skills |
| DocuSign, Ironclad, Definely, Datasite | Transaction tools | Signature status, CLM records, drafting aids, data rooms |
| Harvey | Legal platform | Bridge for firms running both |
Gates, tags and the interview: the controls that matter
Three mechanisms do most of the safety work, and each is written into the open-source YAML rather than promised in marketing.
- [verify] tags. Any citation not checked against a connected research tool is tagged
[verify]in the output. It is a visual reminder that the plugin drafted the citation from the model's memory, which, as the hallucinations guide explains, is exactly the failure mode that has put 811 lawyers into Damien Charlotin's database. - Send and review gates. The litigation plugin will not send a demand letter until a human confirms the FRE 408 settlement-privilege framing; the IP plugin gates takedowns; the corporate plugin pauses board consents for review; legal holds have issue and release gates. These are approval steps the agent cannot skip, the pattern the agent security guide argues for.
- The cold-start interview. Because the profile is a file, it can be version-controlled, reviewed by the head of practice and shared. It is also, as GC AI pointed out in its review of the plugin, single-user Markdown rather than a team playbook, and it resets per session unless you maintain it (GC AI).
Installing and customising in about an hour
- Confirm the plan. Team, Enterprise and API use is never used for training. Team and Enterprise chat interfaces are not eligible for zero data retention as of September 2026, but Claude Code under Enterprise is (Anthropic). On Pro or Max, switch off the model-improvement toggle first; the confidentiality checklist covers each plan.
- Pick the surface. Claude Cowork for lawyers who want a desktop agent on a folder; Claude Code for pipelines and bulk work; Microsoft 365 if the firm wants everything inside Word and Outlook.
- Install one plugin from the marketplace listing in the repository, not all twelve. Commercial-legal is the usual first choice.
- Run the cold-start interview with the head of that practice in the room, and treat the resulting CLAUDE.md as a policy document.
- Attach one research connector before anyone drafts anything citing authority, and one DMS connector once IT has reviewed the permission model.
- Test with synthetic documents, then with a closed matter, then live. Keep every gate on. Only the legal-builder-hub's community skills should ever come from outside the firm, and only after its injection scan and licence gate.
Where it falls short
The honest limits come from Anthropic's own partners. Harvey's LAB benchmark, launched 6 May 2026, scores an agent only if every rubric criterion passes. On that strict metric the best model on 18 September 2026, Claude Fable 5, completes 13.3% of long-horizon legal tasks; Opus 5 scores 11.7% and Opus 4.7, which topped Harvey's headline BigLaw Bench at 90.9%, scores 7.1% (Harvey). Criteria-level pass rates sit in the 80s. Read that as: the drafts are mostly right and rarely entirely right, which is why the repository insists on attorney review.
GC AI's three gaps still apply: no persistent context beyond the CLAUDE.md file you maintain, no citation verification without a connector, and single-user profiles rather than firm-wide playbooks. And the plugins are US-centric in their statutory references (FMLA, NPRM, DMCA §512, FRE 408); a European firm will rewrite the employment and litigation skills before they are usable, which the licence permits. The comparison with OpenAI's Astra for Law, announced 17 September 2026, is the one most firms will now run: open plugins with your own connectors versus a configured model with a 230-million-URL legal index.
What to do this week
Install commercial-legal on one paid seat, run the interview with your standard NDA positions, and put twenty inbound NDAs through /commercial-legal:review against your existing associate markups. Count the deviations found and the compliant clauses it wrongly edited. That single number will tell you more than any webinar, and the NDA review at scale workflow shows what a production version looks like once the number is acceptable. Module 4 of the AI Lab for Lawyers course covers agents, plugins and connectors like these in its custom-systems session, so you can decide which surface fits your firm before you install anything.
Frequently asked questions
What is Claude for Legal?
Claude for Legal is Anthropic's legal offering, launched as a single plugin at the start of February 2026 (announced 30 January–2 February; legal-publisher shares fell on 3 February) and expanded on 12 May 2026 to 12 practice-area plugins and 20 or more connectors. The plugins are open source under Apache 2.0 at github.com/anthropics/claude-for-legal and run inside Claude Cowork, Claude Code, Managed Agents and Microsoft 365 for paid Claude customers at no extra charge.
Which practice areas do the Claude legal plugins cover?
Twelve plugins ship as of September 2026: commercial, corporate, employment, privacy, product, regulatory, AI governance, IP, litigation, legal clinic, law student and a legal builder hub for community skills. Thomson Reuters maintains a thirteenth partner plugin, cocounsel-legal, which requires a CoCounsel subscription and brings Westlaw Deep Research into Claude.
Is Claude for Legal safe for confidential client documents?
It depends on the plan and surface. Team, Enterprise and API usage is never used for training, but Team and Enterprise chat interfaces are not eligible for zero data retention as of September 2026, while Claude Code under Enterprise is. Documents reached through the iManage or NetDocuments connectors stay in the DMS under the lawyer's own permissions. Consumer Pro and Max plans require the training toggle to be switched off.
Does Claude for Legal verify citations?
Only when a research connector is attached. Citations that have not been checked against CourtListener, Trellis or a connected research tool such as CoCounsel are tagged [verify] in the output, and the repository states that every output is a draft for attorney review. Without a connector the plugin is drafting, not researching, and every authority must be opened and read by a lawyer.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Tools
Claude Cowork for Lawyers: Desktop Agents on Your Own Files
Claude Cowork lets Claude read, edit and create files in a folder you choose. What it does for lawyers, the February 2026 deletion incident and a safe setup.
9 min read
Playbooks
Connect Claude to iManage or NetDocuments with MCP: A Setup Guide
Connect Claude to iManage or NetDocuments with MCP in six admin steps, test on a read-only matter, enable audit logging and scope permissions before rollout.
11 min read
Comparisons
Claude for Legal vs Astra for Law: Anthropic and OpenAI Go Direct to Lawyers
Claude for Legal vs Astra for Law: open-source plugins and MCP connectors against a 230m-URL US law index with a ZDR API, data terms, prices, the European view.
11 min read
Guides
What Are AI Agents? A Practical Guide for Lawyers
AI agents for lawyers explained in plain terms: a model plus tools, a loop and a goal. How legal agents review NDAs, where they fail and what to expect in 2026.
10 min read
Glossary
Skills and Plugins (AI) — Definition for Lawyers
A skill is a packaged, reusable instruction an AI tool loads for one task; a plugin bundles skills with tools, connectors and automations. How they differ, and governance.
3 min read
Guides
Agent Security for Lawyers: Prompt Injection, Data Leaks and Guardrails
AI agent security for law firms: prompt injection via hostile documents, the Grok Build and Cowork incidents, and the allowlists and gates that contain them.
11 min read