Claude for Legal: Plugins, Artifacts and a First-Week Pilot

Claude for Legal is worth a supervised pilot when you have a repeatable task, approved source material, and a lawyer who owns the result. Anthropic's May 2026 launch brought 12 practice-area plugins; its October artifacts gallery adds ready-made review tables and other tools that make a first pass easier to inspect. Start with contract triage or a cite check, then decide whether the review burden justifies wider use.

Verified against Anthropic's live product pages, documentation, terms, and embedded contract-review demo on October 30, 2026. The gallery announcement is dated October 28, 2026. This assessment describes the published capabilities and demonstration, not a production trial. May announcement, artifacts announcement.

What Claude for Legal Does Today

Claude for Legal packages legal workflows around Anthropic's Claude assistant. The practical change is that you can start from a defined task and reviewable output, instead of writing a fresh instruction for every contract or research request.

On May 12, 2026, Anthropic announced 12 practice-area plugins and 20+ new MCP connectors. MCP, or Model Context Protocol, is the connection standard that lets Claude work with external systems. The October gallery adds reusable interactive tools. The October announcement says Anthropic has added more than 60 legal connectors since launch; that is the vendor's reported expansion, not a count of services included in your subscription. May release, October update.

The Claude Legal Tool Has Three Layers

Think of the setup as a matter workspace: connectors bring in the files, plugins supply working instructions, and artifacts organize the result. A plugin is a bundle of skills and tools; a skill supplies reusable instructions for a task. An artifact is an editable output, such as an interactive review table or tracker. Legal product overview, plugin documentation.

These layers solve different problems. A well-written playbook cannot compensate for a missing amendment. Access to the right repository does not tell Claude which departures your partner accepts. A polished table makes findings easier to inspect, but its appearance says nothing about their correctness.

The October announcement names Cite check, Jurisdiction survey matrix, Matter board, Playbook deviation board, Regulatory horizon scan, and Tabular review. Their jobs range from flagging citation problems to organizing matters, comparing jurisdictions, checking contracts against approved positions, and tracking regulatory developments. Each opens with an example that you can adapt. Gallery announcement.

For a pilot, choose the output your reviewer can most readily check. An employment-agreement table with passage links is easier to evaluate against a known document set than an open-ended request to find every risk facing the business.

Choose a Claude for Legal Plugin

Start with the practice closest to the work you already repeat. Commercial Legal supports playbook-based contract review; Corporate Legal covers transaction diligence; Litigation Legal includes chronologies and litigation preparation. Anthropic's suite also includes Employment Legal, Privacy Legal, Product Legal, Regulatory Legal, AI Governance Legal, IP Legal, Law Student, Legal Clinic, and Legal Builder Hub. Anthropic's legal plugin repository.

The launch describes a setup interview covering your playbook, escalation chain, risk preferences, and drafting style. Have the supervising lawyer supply those answers. Installing the package is the easy part; deciding which position is standard, which is a fallback, and who can approve an exception is the work that makes it useful. Plugin launch details.

Which Plans Support Claude for Legal Work?

Pro, Max, Team, and Enterprise support plugins; a firm pilot should separately evaluate the commercial data terms and administrative controls. Anthropic's plugin help page states: “Plugins are available to all paid plans (Pro, Max, Team, Enterprise).” Plan availability.

Sources: plugin access, Cowork, artifact feature matrix, plan controls.

The public legal announcement does not provide a template-by-template entitlement table. Verify that the intended legal template and its required capabilities are enabled in your actual workspace before promising a rollout. The artifact documentation also requires the file-creation capability to be enabled, including through organization settings on Team and Enterprise. Artifact requirements.

For a small commercial pilot, Claude's pricing page lists Team Standard at $25 per seat/month billed monthly, with a two-user minimum. Claude pricing.

Claude Cowork for Legal

Claude Cowork carries out tasks across files and connected tools. Anthropic describes a contract workflow that takes a folder of agreements and a playbook, then returns a memo for each contract with identified departures tied to clauses. Its legal page also describes review with tracked changes inside Microsoft Word. Cowork capabilities, legal workflows.

Choose the surface to match the review. A lawyer negotiating wording may want a Word redline. A legal-ops lead assessing a portfolio may want a table. A practice team assembling a chronology may want a document. The pilot succeeds when the reviewer can inspect the result in the form the work requires.

What the Data-Handling Terms Say

Use the terms for the account and configuration you are deploying. Paying for an individual plan does not give it the commercial terms of a managed firm workspace. Anthropic distinguishes Free, Pro, and Max consumer products from its commercial offerings. Commercial data-policy scope.

The legal product page states:

On Team and Enterprise plans, Anthropic doesn’t train on your content by default, and Enterprise admins set how long it’s retained.

The Commercial Terms separately state: “Anthropic may not train models on Customer Content from Services.” They also treat customer content as confidential information, subject to the agreement's provisions and exceptions. Anthropic Commercial Terms.

There are configuration details to resolve before introducing matter material:

• Feedback has a separate policy. Anthropic says explicitly submitted feedback or bug reports, or another opt-in, may be used for training. Team and Enterprise owners can disable the Rate chats setting under Organization settings > Data and Privacy. Decide that setting as part of the pilot. Training and feedback policy.
• Retention needs its own review. Enterprise custom retention is a plan feature, but Anthropic's legal-use guidance also describes model-specific safety retention and narrowly scoped human review of flagged content. Establish which rules apply to your selected model and workflow; do not treat a no-training commitment as a zero-retention promise. Anthropic's configuration guidance.
• Sharing requires a second check. Artifact users connect their own apps, but an artifact can also have shared storage visible to its users. Check both the source-system access and the output's audience before colleagues or clients join. Artifact connections and storage.

Ask the firm's responsible counsel and security owner to approve the configuration against client commitments and matter restrictions. A vendor's product label does not settle whether a particular deployment is suitable for a particular matter.

Which Legal Connectors Does Anthropic Name?

Begin with the system that holds your approved source documents. A connector gives Claude a way to interact with that system; it does not establish that your Claude subscription includes the underlying service.

The following names preserve the wording of Anthropic's May announcement:

• Contract lifecycle and drafting: Definely, Docusign, Ironclad.
• Deal rooms and transaction documents: Box, Datasite.
• Document management: iManage, NetDocuments.
• Expert networks and skills: Lawve AI; The L Suite, with its Lloyd and TopCounsel connectors.
• E-discovery and review: Consilio, Everlaw, Relativity.
• Thomson Reuters: connects Claude to CoCounsel Legal.
• Legal research and case law: Legal Data Hunter, Midpage, Trellis.
• Legal AI assistants: Harvey, Solve Intelligence.
• Public service: BoardWise, Courtroom5, Descrybe, Free Law Project.

Anthropic's connector announcement. The October update additionally names Bloomberg Law, Filevine, Ivo, Legora, and Patlytics. October connector update.

These are named launch and update examples, not an exhaustive current directory. Anthropic's repository identifies connectors that require a customer's own subscription, including a CoCounsel Legal subscription with its MCP connector enabled. Repository access requirements.

For week one, connect the smallest approved source collection that can answer your chosen question. Confirm the pilot user's permissions, permitted actions, and service entitlement. A failed connection is an access problem to resolve, not a reason to improvise an unrestricted document export.

A Worked Example From Anthropic's Contract-Review Demo

The useful unit is a finding you can trace back to its source. Anthropic's legal page embeds a tabular-review demonstration with agreements, questions, extracted answers, source passages, page references, and reasoning. The entries below come from that published sample; they are demonstration material, not client results or a test performed for this article. Anthropic's embedded tabular-review sample.

The sample includes a Quarrendon Markets services agreement and its first amendment. It also includes a Corvanleigh Health Network subscription agreement. Follow the review through the change-of-control question, which asks what happens if the supplier is acquired:

1. Read the agreement finding. For the Quarrendon original, the sample identifies a counterparty right to terminate. Its supporting explanation makes that conditional on the acquirer or an affiliate being a competitor. The review task is to inspect that condition, not stop at the category label.
2. Open the supporting passage. The demo associates the finding with a quoted passage and a page reference. A reviewer should check the full clause, relevant definitions, and surrounding provisions before carrying the finding into a diligence report.
3. Reconcile the amendment. The Quarrendon amendment has “Not found” in its change-of-control field. That is not the same as a conclusion that the agreement family has no restriction. Keep the original and amendment together and have the lawyer establish the operative position.
4. Route a different result. The Corvanleigh sample identifies a consent requirement and classifies the deal risk as high. Its explanation points to language treating a change of control as an assignment. That belongs in the lawyer's review queue, where the transaction facts and complete documents can be considered.

The pilot lesson is specific: test whether the workflow preserves document relationships and uncertainty. A table can make the distinction between a missing answer and an affirmative finding visible. Your review procedure must preserve that distinction through to the final work product. This is an inference from the sample's separate rows, not a claim that Claude cannot analyze amendments.

For a broader comparison of dedicated contract products, use the site's AI contract review software guide. This pilot answers a narrower question: whether Claude's available workflow fits your source material and review process.

Claude for Legal Teams: Who Should Pilot It?

Pilot now when the task is bounded and the reviewer already knows what good work looks like. The best first workflow differs by role.

For a law-firm partner: choose a familiar agreement type or a completed matter with approved materials. Have the supervising lawyer compare the first pass with the issues the matter required. Look for omitted conditions, lost amendments, and explanations that sound stronger than their sources. The decision is whether the workflow helps a lawyer reach a defensible work product with a manageable correction burden.

For in-house counsel: start with departures from an established contract playbook. Supply approved positions and identify who may authorize exceptions. If the organization has not agreed on its fallback positions, settle those before asking Claude to enforce them.

For a legal-ops lead: own the access, routing, and record of review. Confirm which repository is authoritative, who can see an artifact, who handles an exception, and where the approved result is stored. Keep substantive legal approval with the designated lawyer.

Wait on a live-matter pilot if the input data cannot be approved, the required connector is unavailable, or no lawyer has capacity to review the output. You can still inspect the public sample while those issues are resolved. An existing workflow that already meets the need does not require replacement because a new gallery exists.

How to Use Claude for Legal in Week One

Finish the week with a reviewed work product and a decision about continuing. Treat the following as a proposed pilot sequence, not a vendor promise about implementation time.

1. 1

   Define the task and acceptance rule

   Choose one task: a contract question table, a playbook review, or a citation check. Name the supervising lawyer. Write down what must be present in the output, which uncertainties must be flagged, and which actions require approval.

2. 2

   Approve the workspace and source set

   Confirm the account's terms, retention requirements, feedback settings, artifact capabilities, and sharing audience. Use approved documents with a known version history. Keep originals, amendments, schedules, and relevant playbook material together.

3. 3

   Configure one legal workflow

   For a plugin, use Customize > Plugins > Discover, select the approved plugin, and choose Add. Complete its practice setup with the responsible lawyer. For an artifact, open the legal gallery's built-in example first, then adapt the intended template to the approved source set. Verify the template is enabled in your workspace.

4. 4

   Review every pilot output

   Open source passages, check missing findings, and record corrections by type: extraction error, unsupported inference, missing document, or routing mistake. Include the effort to configure, verify, and correct the work when assessing usefulness.

5. 5

   Continue, narrow, or stop

   Continue with the same task if the reviewer can trace the material findings and resolve the corrections consistently. Narrow the document type or question if failures cluster around ambiguity. Stop live use if permissions, provenance, or substantive review cannot be made reliable enough for the intended work.

Installation and template behavior are documented by Anthropic; the pilot sequence and acceptance approach above are editorial recommendations. Plugin installation, legal gallery.

Where a Lawyer Must Stay in the Loop

A lawyer owns the scope, the interpretation, and the final decision. Keep review attached to those points instead of adding a vague reminder at the end of a workflow.

At intake, counsel determines whether the question and source set are appropriate. During review, counsel checks whether the cited material supports the proposition and whether important context is missing. Before release, counsel resolves exceptions and decides what can be communicated or relied on.

For a cite check, Anthropic describes flags for mismatched quotations, incorrect page numbers, and sources that do not support a claim. Treat those as review leads. A flag-free output is not evidence that every relevant authority has been found or that the legal position is sound. Cite-check description.

For contracts, require a reviewer to distinguish quoted wording from interpretation and a missing field from an absent obligation. For shared artifacts, identify who approves a changed template or playbook before others use it. These are operating decisions for the legal team, not tasks solved by an attractive interface.

The broader AI tools for lawyers guide places these choices in a wider legal workflow.

What's Overhyped

Ready-made tools lower the setup barrier; they do not remove the work of configuring and supervising them. The misleading leap is from a neat demonstration to an assumption that the firm's inputs, permissions, and standards will already fit.

A source link helps you inspect an answer. It does not certify the answer. A connector makes a service reachable. It does not include that service's subscription or prove that every relevant document is accessible. A no-training commitment addresses a particular use of data. Retention, feedback, sharing, and matter suitability still need their own decisions.

There is no basis here for promising a universal accuracy rate or a fixed saving in lawyer hours. Measure your own correction burden and review effort on the bounded pilot before expanding it.

The Monday move: the partner names a reviewer, in-house counsel supplies the approved playbook, and legal ops prepares a restricted source set. Open Anthropic's sample together, select one workflow, and make the first decision about reviewed work rather than a demo impression.

Subscribe to the newsletter for practical guides to adopting AI at work.