Your Methods, Your Tools (Skills, MCP and Plugins)

Ask a general AI assistant to review an NDA from Acme against your firm's position, and it runs into two walls. First, it doesn't know your position: which clauses you push back on, which you accept, how you set out your comments. So it falls back on a generic checklist. Second, it can't open the NDA if the file sits in your document system, because it has no way in.

So you download the file, paste it in and type out your position. Then you do it again for the next NDA.

Three add-ons remove those walls. A skill gives the assistant your method. An MCP connection gives it access to a tool or a source. A plugin packages both so a whole team can install them in one step. You don't need to build any of them to use AI well. But they show up in product settings and vendor pitches, and they explain much of what a legal AI tool can actually do with your work.

Skills: your method, written down once

What makes skills practical is when they load. The assistant doesn't read every skill all the time. At the start of a conversation it sees only each skill's name and description, about 100 tokens each. When your request matches a description, it opens that one skill and reads the full instructions Anthropic · Agent Skills docs (opens in a new tab). So a team can install dozens of skills without filling the context window.

A skill has no set size. It can be one sentence: “When you write to a client, sign off with the matter number.” It can also run past a thousand lines, with every clause of your NDA position and the reasoning behind it. Both are real skills.

Long skills work because the same rule applies inside the folder. Anthropic advises keeping the main file under 500 lines and moving the detail into separate files that the main file points to Anthropic · Skill authoring best practices (opens in a new tab). The assistant opens those files only when a step needs them, and it runs a program without reading it, so only the result enters the conversation. A long playbook takes no room in the conversation until a task needs it.

A skill is not the same as a prompt. A prompt is what you type in one conversation. It works for one task. When the conversation ends, the prompt is gone, and next time you must type it again. A skill stays saved. The model loads it only when the task needs it, with its files. You write the guidance one time and don't repeat it in each conversation.

Standing instructions, such as a project's custom instructions, are saved too, but they apply to every message whether they are relevant or not. Chapter 6 shows how to turn your own methods into skills.

One caution. A skill can tell the assistant to do anything the assistant is allowed to do, including run the programs inside it. Anthropic's own advice is to use skills only from sources you trust and to read every file before installing one. A skill is plain text, so you can read it, and you should.

MCP: one plug for many tools

Lesson 1.2 showed that an agent can only act through the tools its harness gives it. Until recently, every AI product had to build its own link to every outside service: one for Outlook, another for a document system, another for a case-law database. Each pairing was a separate project, so most were never built.

The Model Context Protocol (MCP) is a shared standard for those links. A service builds one MCP connection, and every assistant that speaks MCP can use it.

Anthropic released MCP in November 2024. Within a year it had been adopted by ChatGPT, Gemini, Microsoft Copilot and others, and counted more than 10,000 public connections. In December 2025 Anthropic handed it to a foundation run under the Linux Foundation, a non-profit, so no single company controls it Anthropic · Donating MCP, Dec 2025 (opens in a new tab).

In products you rarely see the letters MCP. Claude and Le Chat call these links connectors; ChatGPT calls them apps. Legal software is arriving fast. In May 2026, Anthropic announced more than 20 connectors for legal work, including the document systems iManage and NetDocuments, the e-discovery platforms Relativity and Everlaw, DocuSign, and Thomson Reuters LawSites · May 2026 (opens in a new tab). Lawve, which publishes this guide, was one of them Anthropic · Claude directory (opens in a new tab).

OpenAI followed in September with Astra for Law, a version of its GPT-6 Astra model set up for legal work, with its own legal search index. It came with 26 legal plugins from vendors such as Thomson Reuters, Harvey and iManage, 9 built by lawyers and legal engineers, and 47 skills LawSites · Sep 2026 (opens in a new tab). Lawve is one of those plugins too OpenAI · Lawve plugin (opens in a new tab).

Each connection is a door, so the three questions from lesson 1.2 apply: what it lets the assistant see, change and send. Well-built connections use your own access rights: if you can't open a matter in the document system, neither can the assistant.

Two more points. What the assistant reads through a connection goes to the model's provider, like anything you type, so the confidentiality rules in Chapter 3 apply. And text that comes in through a connection, such as an email or a document from the other side, can contain hidden instructions. That risk, prompt injection, is also covered in Chapter 3.

Plugins: the package a team installs

Skills and connections work on their own. A plugin bundles several of them, often with ready-made commands, so they install in one step and everyone gets the same setup. A contracts plugin might bring a skill for NDAs, a skill for vendor agreements, the connection to your document system and a command that starts a review.

The idea is spreading across products. OpenAI describes a plugin as “a reusable package for a workflow that can include skills, apps, or both”, and is moving custom GPTs into plugins OpenAI · Migrating custom GPTs (opens in a new tab).

Plugins are also how legal AI made the financial news this year. In early February 2026, Anthropic released a legal plugin for Claude Cowork, its desktop agent, covering contract review, NDA triage and compliance work, set up around an organisation's own playbook. On 3 February, shares in RELX, Wolters Kluwer and Thomson Reuters, the groups behind LexisNexis, Kluwer and Westlaw, fell by between 10% and 14% Canadian Lawyer · Feb 2026 (opens in a new tab).

The plugin itself was mostly instructions written in plain language, plus connections. That is the lesson for buyers: much of what makes an AI tool “legal” is skills and connections around a general model. In May, Anthropic followed with 12 plugins for areas such as commercial, employment, privacy, IP and litigation work, each starting with a setup interview about the team's playbook, escalation rules and house style. All twelve are published on Lawve, where you can see the skills and commands each one bundles before you install it. Chapter 4 looks at what that means when you choose between a legal AI tool and a general assistant.

How the pieces fit

Back to the Acme NDA, with the contracts plugin installed. You type “Review the Acme NDA against our position.” The assistant matches the request to the NDA skill and reads your checklist. It uses the document system connection to find and open the file. The model then does what lesson 1.1 described: it reads, compares and writes, in the format your skill asks for. You review the result.

The skill supplied the method, the connection supplied the file, the plugin delivered both, and the model did the reading and writing.

What this means for your work

• You don't have to build anything to benefit. Many tools already ship with skills and connections. Ask which ones, who wrote them and who keeps them current.
• Your know-how can become something the team installs. A checklist that lives in a partner's head or an old Word file can become a skill everyone uses the same way. Chapter 5 covers who should maintain it; Chapter 6 shows how to write it.
• Treat each add-on like software. Read the skills, check what each connection can see, change and send, and know what a plugin bundles before you install it.

Next, lesson 1.4 turns from how these tools work to what lawyers are actually using them for, with the figures from recent surveys.