Updated 9 min read
Claude for Legal: What It Does, Where It Stops, and How to Connect Your Contracts

Claude for Legal: What It Does, Where It Stops, and How to Connect Your Contracts

Claude for Legal is Anthropic's package of legal workflows for its Claude AI models. It comes in two forms: a Legal plugin for Claude Cowork, and an open-source repository on GitHub, released under the Apache 2.0 licence, with plugins organised by practice area, agent skills and connectors to outside systems.

The repository has 12 practice-area plugins. They cover commercial work (vendor agreements, NDAs, SaaS subscriptions, renewals), corporate work (diligence, closing checklists, board consents), privacy (DPA review, data subject requests), product, employment, AI governance, regulatory, IP and litigation, plus plugins for law students, legal clinics and a builder hub for community skills. The README is explicit about the limit: "Every output from these plugins is a draft for attorney review, not legal advice, not a legal conclusion, not a substitute for a lawyer."

The three names people search for

The terms get used interchangeably, so here is what each one means:

  • The Claude legal plugin usually means Anthropic's Legal plugin in the knowledge-work-plugins collection: an add-on that gives Claude in-house legal commands such as contract review and NDA triage.
  • The Claude Cowork legal plugin is the same plugin installed in Claude Cowork, Anthropic's agentic desktop app. It also runs in Claude Code.
  • Claude legal skills are the individual packaged instructions inside these plugins, each teaching Claude one task, such as reviewing a vendor agreement or answering a data subject request. Skills from the open-source repository can also run through the Claude Managed Agents API inside a company's own workflow tools.
Claude for Legal at a glance
Made by
Anthropic
Where it runs
Claude Cowork, Claude Code, or the Claude Managed Agents API
Licence
Apache 2.0 (the repository)
Contract commands
/review-contract, /triage-nda, /vendor-check
Connects to other systems
Through MCP connectors
Anthropic's caveat
Every output is a draft for attorney review

If you searched for "Claude for legal" because a colleague said legal tech is finished, this guide is the calmer version: what the plugins actually do, the one thing they cannot do on their own, and how to close that gap.

The Claude Cowork legal plugin lists five commands. Three of them are contract work:

/review-contract
Reviews a contract clause by clause against the negotiation playbook you configure, flags each issue green, yellow or red, and suggests specific redlines.
/triage-nda
Sorts incoming NDAs into standard approval, counsel review or full review, so routine ones stop queueing behind complex ones.
/vendor-check
Checks the status of agreements with a vendor, which only works if Claude can see where those agreements live.
/brief and /respond
Briefings (daily briefs, topic research, incident response) and templated replies for things like data subject requests and discovery holds.

You set your positions in a playbook file called legal.local.md, saved in a folder you share with Cowork. The plugin README warns that its default playbook examples reflect US positions (Delaware, New York, California), so teams under EU, UK or other law need to rewrite them before relying on the analysis.

The open-source repository goes further than the Cowork listing. Its commercial plugin starts with a cold-start interview that captures your positions, sales side, purchasing side or both, and its review command produces what the README calls a "deviation-by-deviation memo against your playbook, with specific redline language and named approver". It also has commands for escalations and amendment history, and a scheduled renewal watcher that posts what is coming up in the next 90 days.

This is genuinely useful work, and it is the same work many legal AI products charge for. The important design choice is the playbook. Claude does not arrive knowing your fallback positions, your approval thresholds or which clauses your CFO cares about. You write them down, and the review is only as good as what you wrote. Our guide to evaluating a legal AI agent covers the questions to ask of any tool doing this.

Where Claude stops: it cannot see your contracts

A plugin that reviews the contract in front of it is half of the job. The other half is everything around that contract:

  • Context. Is this a renewal? What did we agree last time? Is there a master agreement this order form sits under?
  • Filing. Where does the reviewed and signed version go, and who can find it next year?
  • Status. Which agreements with this vendor are signed, which are still in negotiation, and when does each one end?

Claude learns none of this from the file you drop into a chat. Anthropic's own README for the commercial plugin says what changes once a contract system is connected: "reviews check for prior agreements with the same counterparty, bulk-load the renewal register, create records with review memos attached." That is why the Claude for Legal repository is built around connectors.

Connectors use the Model Context Protocol (MCP), an open standard for giving AI models access to outside tools. The repository ships connectors for some contract and document systems, including Ironclad and iManage, along with general tools such as Slack, Google Drive and Box. Connectors for paid systems need your own subscription to that system. Anthropic's connector guide says the plugins have a fallback for when a connector is not responding, so they still run, just without that context.

For one lawyer reviewing one contract, it is a capable start once the playbook is written. For a team that owns every contract the company signs, it covers the review and leaves the rest of the lifecycle to something else.

What it does well
Clause-by-clause review against your written playbook, NDA triage, vendor status checks when a system is connected, briefings and templated replies.
No shared contract library
It reads the file you give it. Signed contracts, versions and who can see what live somewhere else.
No metadata or renewals of its own
Counterparty, value, end date and notice period are not stored by the plugin, so renewal alerts depend on a connected register.
No negotiation with the other side
Redlines come back to you. Sending a round to the counterparty, reading their changes and tracking the next round happen outside it.
No e-signature
Signing and storing the executed version need another tool.
No team-wide audit trail
There is no shared record of who reviewed, approved or changed what across the legal team.

The plugin's own design points to the answer: connect it to the system where your contracts live. That system should have an MCP connector or a documented API, and it should give Claude its own scoped access rather than a person's full login. If it does, Claude handles the questions and drafts, and the contract system keeps the record.

The question to ask your contract system

Not "do you have AI?" but "can my AI reach my contracts safely?" That means a documented API or connector, credentials for each AI tool separately, and access limited to the folders the task needs.

What is the best Claude connector for managing company contracts?

The best connector is the one that reaches the system your contracts already live in, so start there. If that is Ironclad or iManage, the Claude for Legal repository has a connector for it.

If your contracts sit in shared folders, or you are choosing a contract system now, look for one with an MCP server Claude can connect to and a documented API for the jobs MCP does not cover. Bind, an agentic AI workspace for in-house legal teams, is one option: Claude connects to Bind MCP as a custom connector rather than through the repository, and the Bind API handles filing documents and updating fields. The Bind section below shows how that works.

How to connect Claude to a contract system that has no connector

You do not need a ready-made connector for Claude to work with a system. You need a documented API.

Claude Code, Anthropic's agentic coding tool, can make HTTP requests and read API documentation. Give it the system's OpenAPI specification and a key, and it can list, read and file documents through the API, or write a small script that does so on a schedule. Developers can also wrap any REST API in their own MCP server; the repository's connector guide describes what a good legal MCP server looks like.

Whichever route you take, the safety rules are the same:

  1. A separate key for Claude, never a person's login
  2. Access to the folders the task needs, nothing else
  3. Read only, unless it must file or update
  4. An expiry date while you experiment
  5. Revoke instantly if anything looks wrong

The point of the scoping is not that Claude is uniquely risky. It is that an agent acting on a loosely worded instruction should not be able to read the board minutes because someone asked it to "check all our agreements".

Our contract management API guide has the full checklist for judging whether a contract system's API is fit for this.

These are not either-or, but they solve different problems.

QuestionClaude for Legal pluginsAgent inside your CLM
Where the contracts liveElsewhere, reached by connectorsIn the same system
Who maintains the playbookYou, in plugin configurationYou, in the CLM
Negotiation rounds and signingOutside the pluginIn the same workflow
Best forTeams already living in ClaudeTeams that want one place for contracts

If your team already works in Claude every day, the plugins are a sensible way to add structured legal workflows, and connecting them to your contract system is what makes them useful past the first document. If you want contracts drafted, negotiated over several rounds, signed and tracked in one place, an agent built into the contract system does that end to end. Many teams will end up using both.

How to do this in Bind

Claude reaches Bind in two ways. Bind MCP, Bind's MCP server, lets Claude, ChatGPT or another AI tool that supports remote MCP connectors with sign-in search and read your contracts in Bind. Ask Claude "Which of our Ottermead supplier agreements are signed, and when do they end?" and it answers from your Bind spaces with a link to each contract. It is read only, signs in with each person's own Bind login, and an organisation admin turns it on. To send documents or update field values, there is the Bind API, a REST API with public documentation, with its OpenAPI specification at help.bindlegal.com/developers. Both are included in every plan at no extra cost.

The questions Claude asks through Bind MCP are the same kind you can ask inside Bind's own chat. Here is Bind answering questions about a contract library:

Asking Bind about your contracts

Connect Claude to Bind MCP in three steps

Connecting Claude to Bind MCP takes a few minutes:

  1. An organisation admin turns it on. In Organization settings → API and MCP access, under MCP access, turn on Allow members to connect AI tools through MCP. It is off until an admin does this, and turning it off disconnects every tool within a minute.
  2. Add Bind in Claude. Bind is in Claude's connector directory: open it and connect. You can also add Bind as a custom connector under Settings → Connectors by pasting the connector address: https://mcp.app.bindlegal.com/mcp, or https://mcp.us.bindlegal.com/mcp if you use Bind at us.bindlegal.com.
  3. Sign in to Bind and choose the organisation. Claude then sees only the documents you can open in that organisation. Every document it reads is recorded in Bind like any other access: who, when and through which tool.

Let Claude file documents through the Bind API

To let an agent file documents or update fields, use the API instead. The steps below set it up for Claude Code, or any agent that can make HTTP requests.

Here is what it looks like from Bind's side when an outside system files a contract through the API, sets a field, and then tries a space it was never given.

  1. Your system sends the contract
  2. It lands in the space it was granted
  3. Fields set from your system
  4. Other spaces stay out of reach

To give Claude access to one part of your contract library:

  1. Create an API client for Claude. In Organization settings → API and MCP access → New API client, name it for the job, for example "Claude vendor checks", so it shows up separately from every other integration.
  2. Give it only the spaces it needs. Grant, say, the "Supplier agreements" space with Read access. Every other space returns 404 to this client, as if it did not exist.
  3. Create a key with an expiry. Choose 30 or 90 days while you experiment. The key is shown once; store it where Claude Code reads its environment, not in a prompt.
  4. Point Claude at the spec. Give Claude Code the OpenAPI document at help.bindlegal.com/developers/openapi.json and ask it to list the documents in the space with their fields and status. From there it can answer "which agreements with this vendor are signed, and when do they end?" from your real data.
  5. Add write access only when you need it. If Claude should file documents, switch the space to Read and write and have it set external_reference to your own ID so a retry never files the same contract twice.

Review, negotiate and sign with Bind's own agent

Bind also has its own AI agent built in. In a Bind chat, /review checks a contract against your playbooks and returns tracked changes with comments, and the same contract can go through negotiation rounds and e-signature without leaving the space. The API is how the rest of your tools, Claude included, take part in that.

For the bigger picture, Aku Pöllänen, Bind's CEO, explains how Bind handles contracts from draft to signature:

See how Bind works

Bind itself is ISO 27001 certified, SOC 2 Type I compliant, and GDPR compliant. It is used by in-house legal teams at companies including Atria, listed on Nasdaq Helsinki, and Outdoor Holding, listed on Nasdaq in the US.

Trusted by fast-moving legal teams around the world

See how Bind would work for your team.

Book a demo
  • AirLife logo: a North American medical device manufacturer and a Bind customer
  • Algol logo: a Bind customer
  • Atria logo: a company listed on Nasdaq Helsinki and a Bind customer
  • OLA Vacations logo: a Bind customer
  • Outdoor Holding logo: a company listed on Nasdaq in the US and a Bind customer
  • Ren-Gas logo: a Finnish green hydrogen and power-to-gas developer and a Bind customer
  • Slush logo: the global startup and tech event organizer and a Bind customer
  • Suomen Jääkiekkoliitto logo: the Finnish Ice Hockey Association and a Bind customer
  • Verda logo: the Finnish AI cloud company and a Bind customer
  • Weiss Technik logo: the environmental test equipment maker in the Schunk Group and a Bind customer

Frequently asked questions

What is Claude for Legal?

Claude for Legal is Anthropic's set of legal plugins, skills and connectors for its Claude AI models, published as an open-source repository under the Apache 2.0 licence, alongside a Legal plugin for Claude Cowork. The repository has 12 practice-area plugins, such as commercial, corporate, privacy, employment, product and litigation work, and includes contract review against a negotiation playbook and NDA triage. Anthropic states that every output is a draft for attorney review, not legal advice.

What is the Claude legal plugin?

The Claude legal plugin is Anthropic's Legal plugin for Claude Cowork, its agentic desktop app, which also works in Claude Code. It adds commands for in-house legal work: /review-contract checks a contract clause by clause against your playbook with green, yellow and red flags, /triage-nda pre-screens NDAs, /vendor-check looks up the status of agreements with a vendor, /brief prepares legal briefings and /respond drafts templated replies. You configure your positions in a playbook file called legal.local.md; the default examples reflect US positions.

Is the Claude Cowork legal plugin enough for an in-house legal team?

For reviewing a single contract or triaging NDAs, it is a capable start once you write your playbook. It is not a contract system: it has no shared repository of signed contracts, no metadata or renewal dates of its own, no multi-round negotiation with the counterparty, no e-signature and no audit trail across the team. Anthropic's own commercial plugin says it does more once a contract system is connected, so most teams pair it with a CLM that Claude can reach through an MCP connector or an API.

What is the best Claude connector for managing company contracts?

The best connector is the one that reaches the system your contracts already live in. The Claude for Legal repository has connectors for systems such as Ironclad and iManage. If your contracts sit in shared folders, or you are choosing a contract system now, look for one with an MCP server Claude can connect to and a documented API. Bind, an agentic AI workspace for in-house legal teams, is one option: Bind is in Claude's connector directory, Claude connects to Bind MCP to search and read your contracts, and the Bind API files documents and updates fields.

What are Claude legal skills?

Claude legal skills are packaged instructions that teach Claude a specific legal task, such as reviewing a vendor agreement, triaging an NDA or responding to a data subject request. Anthropic publishes them inside its legal plugins, and the open-source Claude for Legal repository includes a legal-builder-hub plugin for finding, installing and checking community legal skills.

Can Claude review contracts?

Yes. Anthropic's Legal plugin includes a /review-contract command that reviews a contract clause by clause against a negotiation playbook you configure, flags each issue as green, yellow or red, and suggests redlines. A separate /triage-nda command sorts incoming NDAs into standard approval, counsel review or full review. The quality of the output depends heavily on the playbook you give it, and a lawyer should review the result.

Does Claude for Legal connect to contract management software?

Through MCP connectors. The open-source Claude for Legal repository ships connectors for some contract and document systems, including Ironclad and iManage, plus general tools such as Slack, Google Drive and Box. Connectors that need the customer subscription use the customer’s own account and API key. Systems without a connector can still be reached by Claude through their API, for example from Claude Code, if the system has a documented API and the key is scoped appropriately.

Is Claude safe to use with confidential contracts?

That depends on how you use it, not only on the model. Check which Claude plan and data terms apply to your organisation, keep contracts in a system with proper access controls rather than pasting them into chats, and when Claude connects to a contract system, give it its own credentials limited to the folders the task needs, read only unless it must write. Anthropic advises lawyer review of all legal outputs.

Does Bind work with Claude?

Yes. Bind MCP, Bind’s MCP server, lets Claude, ChatGPT or another AI tool that supports remote MCP connectors with sign-in search and read your contracts in Bind. It signs in with each person’s own Bind login and sees only what they can open in Bind; an organisation admin turns it on. To send documents or update field values, Claude Code can use the Bind API, a REST API with public documentation and an OpenAPI specification at help.bindlegal.com/developers, with an API key that an admin limits to specific Bind spaces. Both are included in every plan at no extra cost. Bind also has its own built-in AI agent for drafting, reviewing, negotiating and signing contracts against your playbooks.