Claude for Legal: Complete Setup Guide for Lawyers (2026)

What Is Claude for Legal?

Claude for Legal is Anthropic's official open-source legal toolkit. Anthropic announced its practice-area plugins and connectors on May 12, 2026, building on a first legal plugin released earlier that year. It bundles 12 practice-area plugins, more than 80 named workflow agents (per the project README), and 20+ connectors across legal research, document management, e-discovery, contract lifecycle, and deal-room platforms, all running through Claude's desktop agent (Cowork) or Claude Code. Its plugins attach source attribution to citations, surface the jurisdiction assumptions they make, and build in gates before anything is filed, sent, or relied on, and every output is a draft for attorney review.

It is not a separate subscription. The toolkit is open source under the Apache 2.0 license, and plugins are available on all paid Claude plans (Pro, Max, Team, and Enterprise), so it layers on top of your existing plan. Some connectors require your own subscription to the connected service. Think of it as switching Claude from general assistant to legal associate.

Why Claude for Legal Matters for Law Firms

General-purpose AI produces general-purpose output. Claude for Legal fixes three problems attorneys hit with base Claude:

• Citation handling: The plugins tag citations pulled through a connected research tool with their source and flag citations drawn from model knowledge alone for verification (per the Claude for Legal README), which supports the verification step that guards against the hallucination risk that has led to sanctions against unprepared attorneys.
• Document analysis at scale: Upload contracts, depositions, or discovery sets running hundreds of pages. Ask Claude to cite the page, paragraph, or clause behind each answer so you can check it against the source.
• Jurisdiction awareness: Each plugin starts with a setup interview that learns your practice, playbook, and house style (see Anthropic's launch post), and the plugins surface the jurisdiction assumptions they make (per the Claude for Legal README) so you can check them. Confirm local rules and procedural requirements yourself.

Clio's 2024 Legal Trends Report found that 79% of legal professionals use AI in some form, up from 19% the year before. Claude for Legal exists because legal work demands precision that generic prompts cannot deliver.

Step-by-Step Setup: Account to First Prompt

Step 1: Choose Your Plan

Claude for Legal runs on any paid Claude plan, since plugins are available on Pro, Max, Team, and Enterprise. Pro ($20/month, or $17/month billed annually) is an individual consumer plan on which model training is opt-out, so use it to try the toolkit on non-client material; firms handling client data should run it on Team (standard seats $20/user/month billed annually, $25 billed monthly) or Enterprise (self-serve at $20/seat/month billed annually plus usage at API rates, or sales-assisted) for the commercial data terms, per Claude's pricing page. For a full plan comparison, see our Claude AI pricing guide.

Choose Team if you are a small firm handling client data; Team plans require at least two members and are not intended for individual use. That leaves solo attorneys without a one-seat commercial option in the Claude apps: Pro and Max are consumer plans, so a solo practitioner who wants commercial data terms for client work needs Enterprise or a Team plan with at least two members. Choose Enterprise if your firm requires SCIM provisioning, audit logs, or custom data retention. SSO is available on both Team and Enterprise, and Anthropic's SOC 2 Type II certification covers its commercial products, not only Enterprise. The 2025 Thomson Reuters Generative AI in Professional Services Report found that only 41% of law firms have established policies governing generative AI use. Whatever plan you choose, pair it with a written AI policy, and choose Enterprise if you need audit logs or SCIM provisioning.

Step 2: Create Your Workspace

Sign in at claude.ai with your firm email and select the Team plan, or upgrade an existing account at claude.ai/upgrade (see Anthropic's Team setup guide). Name the organization after your firm and add attorneys as members. Chats stay private to the member who created them unless shared, even inside a shared project (per Anthropic's help center), but the Primary Owner can export the organization's data, including conversations.

Step 3: Install Claude for Legal

In Claude's desktop agent (Cowork), install plugins from Customize > Browse plugins; in Claude Code, add the repository as a plugin marketplace and use /plugin install (per the Claude for Legal README). The toolkit is open source, so you can review exactly what each component does. You will make three choices:

• Practice-area plugins: Pick from the 12 plugins (commercial, corporate, employment, privacy, product, regulatory, AI governance, IP, and litigation, plus law student, legal clinic, and builder-hub plugins). This tunes Claude's output for your work.
• Workflow agents: The plugins you install bring their named workflow agents with them (80+ in all), such as the Vendor Agreement Reviewer, NDA Triager, and Deposition Prep agents. Skills fire automatically when relevant, and slash commands run specific workflows. Each plugin's setup interview asks about your practice, so give it your primary jurisdiction and house style.
• Connectors: Connect the major legal research and document platforms your firm already uses, so Claude can read from them and, where the connector is scoped for it, write to them.

Step 4: Configure Security and Confidentiality

This step is non-negotiable. Start with the plan, then the settings:

• Run client work on a Team or Enterprise plan. Under Anthropic's commercial terms, your data is not used for model training by default and Anthropic acts as a data processor (see Anthropic's privacy center). Custom data retention controls are Enterprise-only (see Claude pricing). This is the baseline for ABA-compliant AI use.
• On Enterprise, enable audit logs if your firm tracks who uses AI and when (audit logs are not included on Team).
• On Team or Enterprise, configure SSO so attorneys sign in through your firm's identity provider.

ABA Formal Opinion 512 (July 29, 2024) reminds lawyers that they must make "reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of the client." The opinion says lawyers should read and understand the terms of use and privacy policy of any generative AI tool they use, and that informed client consent is required before entering client information into a self-learning tool whose output could disclose it. Commercial plans with no-training-by-default terms address a core risk the opinion identifies, but they do not end the analysis: review the terms and decide whether client consent is needed. Consumer plans (Free, Pro, and Max) run under consumer terms that leave model training to an opt-out rather than excluding it by default, so keep client information off them; Opinion 512 requires you to evaluate the risk of disclosure before putting information relating to a client's representation into a generative AI tool.

Your First Three Legal Prompts

Once Claude for Legal is active, try these three prompts to see what it can do. Each uses the CRAFT framework for optimal results.

Prompt 1: Contract Red-Flag Scan

Upload a contract and use this prompt:

"Review this agreement as a [practice area] attorney in [state]. Identify all clauses that create risk for [party]. Flag any non-standard terms, missing protections, or provisions that deviate from market norms. Format your findings as a table with columns: Clause, Section, Risk Level (High/Medium/Low), and Recommended Action."

The prompt asks for a structured table with section references; check each reference against the contract. In a 2018 study run by the legal AI vendor LawGeex, 20 experienced attorneys averaged 92 minutes to review five NDAs while its AI completed the same task in 26 seconds with higher average accuracy (94% versus 85%), according to LawGeex's NDA review benchmark.

Prompt 2: Research Memo Draft

"Draft a research memorandum on [legal question] under [state] law. Include: (1) a brief answer, (2) the governing statutory framework, (3) key case law with full citations, (4) analysis of how the law applies to [specific facts], and (5) a conclusion with recommended next steps. Use Bluebook citation format."

Citations that come through a connected research tool are tagged with their source, and citations from model knowledge alone are flagged for verification, but the plugins do not guarantee accuracy. You still verify every citation, but the starting point is far better than a general AI draft.

Prompt 3: Deposition Summary

"Summarize this deposition transcript. For each topic discussed, provide: the topic heading, page-line references, key admissions or contradictions, and any objections made. Highlight testimony that supports or undermines [specific claim]."

For a deeper dive on deposition workflows, see our deposition summary guide.

How Claude for Legal Compares to Harvey AI and CoCounsel

Here is how Claude for Legal compares with two widely used legal AI platforms:

The 8am 2026 Legal Industry Report found that 69% of legal professionals now use generative AI tools for work, up from 31% the year before, while only 46% of firms have implemented general-purpose AI tools. Claude's self-serve model and lower price point make it a natural fit for solo and small firm attorneys. Harvey, which says more than 80 Am Law 100 firms use it, and CoCounsel remain strong choices for larger firms with existing Westlaw or enterprise AI budgets.

Security Best Practices for Law Firms

Setup is only half the job. Maintain these practices to stay compliant:

• Written AI policy: Every firm needs one. Define which plans are approved, what data can be entered, and how output must be verified.
• Client disclosure: ABA Formal Opinion 512 says whether you must tell clients about AI use depends on the facts, but you must disclose it if a client asks or if the engagement terms or outside counsel guidelines require it, and you need informed consent before client information goes into a self-learning tool. State and local bars may add their own guidance, so check your jurisdiction.
• Prompt hygiene: Never paste opposing counsel's privileged communications into Claude. Redact sensitive identifiers when possible.
• Output verification: Treat every Claude response as a first draft. Verify citations, check calculations, and confirm factual claims. The plugin helps, but final responsibility stays with you.