Claude for Legal: Open-Source Plugins for Attorney-Supervised Legal Workflows

What Claude for Legal Is and Who It Is For

Claude for Legal is a reference implementation: a collection of plugins, agents, and connectors that Anthropic has built for the legal workflows it sees most commonly. The repository covers in-house commercial, privacy, product, corporate, employment, litigation, regulatory, AI governance, IP, and academic legal work (law school clinics and students).

The intended users are two groups. First, legal departments that already use Claude Code or Claude Cowork and want production-ready starting points for specific workflows rather than building from scratch. Second, organizations that want to deploy legal agents programmatically through the Claude Managed Agents API behind their own workflow engine.

A critical constraint is stated prominently in the README: every output is a draft for attorney review, not legal advice, not a legal conclusion, and not a substitute for a lawyer. The plugins are built with source attribution on every citation, conservative defaults on privilege and subjective legal calls, jurisdiction assumptions surfaced in the output, and explicit gates before anything is filed, sent, or relied on. The attorney using the plugin, not the plugin, is responsible for the legal positions taken in work product.

Two Deployment Modes from One Codebase

The repository is designed to run from the same prompts and skills in two modes. The first is as a Claude Cowork or Claude Code plugin, installed on a local machine and invoked through the Claude interface. The QUICKSTART.md file describes an installation path that takes about 60 seconds. The README links to a video showing the Cowork installation flow.

The second mode is deployment through the Claude Managed Agents API. This allows the plugins to run behind an organization's own workflow engine: a scheduled job, a legal operations platform, or a custom application. The same system prompt and the same skills execute in either mode; the deployment surface is the only thing that changes.

The repository top level reflects this structure: .claude-plugin/ holds the plugin configuration, and managed-agent-cookbooks/ holds recipes for the scheduled, eyes-on-the-feed workflows such as the Renewal Watcher, Docket Watcher, Regulatory Feed Monitor, and Diligence Grid.

Practice Area Plugins and Named Agents

Each practice area has its own directory: commercial-legal/, corporate-legal/, employment-legal/, privacy-legal/, ip-legal/, litigation-legal/, regulatory-legal/, ai-governance-legal/, product-legal/, law-student/, and legal-clinic/. Each plugin is built around a cold-start interview that learns the organization's playbook, and a CLAUDE.md practice profile that every skill in that practice area reads from.

Named agents map specific workflows to commands:

The Vendor Agreement Reviewer runs as /commercial-legal:review and reviews a vendor MSA against the playbook, producing a redline memo. The NDA Triager also runs as /commercial-legal:review and produces a GREEN/YELLOW/RED triage of inbound NDAs. The Amendment Tracer runs as /commercial-legal:amendment-history and traces how a contract has changed across its base agreement and amendments.

In employment, the Termination Reviewer (/employment-legal:termination-review) runs a proposed termination against jurisdiction-specific risk flags. The Worker Classification Screener (/employment-legal:worker-classification) tests a proposed engagement against the controlling state test.

In corporate work, the Tabular Diligence Review (/corporate-legal:tabular-review) produces a tabular review of a data room with one row per document and every cell cited. The Board Consent Drafter (/corporate-legal:written-consent) drafts unanimous written consents in house format with precedent search.

MCP Connectors: General and Legal-Specific Systems

The repository includes MCP connectors documented in CONNECTORS.md. General productivity connectors cover Slack, Google Drive, and Box. Legal-specific connectors cover Ironclad (contract lifecycle management), DocuSign (electronic signature), iManage (document management), Everlaw (litigation support), and CourtListener (free law and court documents).

These connectors allow the agents to read documents from and write back to the systems a legal team already uses, rather than requiring manual file uploads. An attorney can invoke the Vendor Agreement Reviewer against a contract stored in iManage directly through the Ironclad or Box connection, depending on where the contract lives.

The external_plugins/ directory holds connectors to additional systems beyond the ones documented in CONNECTORS.md. The external_plugins/ directory is listed in the top-level file structure but the README does not document its contents in detail.

Limitations: Jurisdiction Assumptions, Evolving Law, and Attorney Responsibility

The README is candid about the limitations. It states: 'The law in many of these areas is unsettled and evolving. The attorney using the plugin, not the plugin, and not Anthropic, is responsible for the legal positions taken in their work product.' This is not a disclaimer added as an afterthought; the plugins are designed with it as a structural constraint.

Jurisdiction assumptions are surfaced in the output rather than silently applied. This means a review memo will flag when it has assumed a particular jurisdiction rather than applying logic for the jurisdiction of the actual contract. The attorney must verify that the assumption is correct.

The plugins do not represent Anthropic's legal positions. Where a skill includes a checklist item, a risk flag, or a characterization of case law, that is an aid to the reviewing attorney's own analysis, not a statement from Anthropic about what the law requires.

Gaps in the workflow are expected: the repo covers the ten most common practice areas but does not claim to cover every legal task. Tax law, patent prosecution, and estate planning are examples of areas not listed in the README.

Comparison with Harvey and Maintenance

Harvey is a widely known commercial AI legal platform targeting law firms and legal departments. The difference in approach is substantial. Harvey is a closed commercial product with its own model fine-tuned on legal data and managed security controls; accessing it requires a commercial contract.

Claude for Legal is Apache-2.0 licensed open-source software. Teams can inspect every prompt, modify the practice profile for their own playbook, and extend the workflow without asking for a product change. The tradeoff is that Claude for Legal requires an organization with the technical capacity to deploy and maintain a Claude Code or Managed Agents integration, while Harvey is a managed service.

The last push was on 2026-09-24. The repository has no GitHub releases. The license is Apache-2.0. The legal-builder-hub/ directory provides tooling for teams that want to extend the repository with custom practice-area plugins beyond the ten included areas.

Editorial conclusion

Claude for Legal is the right choice for in-house legal teams and law firms that already use Claude Code or Claude Cowork and want to accelerate review, triage, and drafting workflows without building the prompt engineering themselves. It is not a legal advice system and should not be treated as one; the README explicitly requires attorney review of every output. Before deploying to production, read the CLAUDE.md practice profile for your area and verify that the jurisdiction assumptions surfaced by each skill match your actual practice.

Frequently asked questions

Official sources

1. anthropics/claude-for-legal on GitHub
2. Issues
3. License: Apache-2.0
4. README