Claude Cowork Plugins and Custom Skills for Teams
Claude Cowork plugins and custom skills built around your team's templates, documents and tools. We turn how your people write proposals, decks, reports and client emails into skills Claude follows, package them with the connectors they need and roll them out across your organization.
Claude Cowork Skills and Plugins in Plain Words
A skill teaches Claude how your team does one task. A plugin bundles skills with their connectors and commands, so one install sets up a whole workflow.
Anthropic built Claude Cowork to bring Claude Code's agentic abilities to knowledge work. You describe the outcome, and Claude works through the files and steps to produce it.
A skill is a folder with a SKILL.md file. Claude reads its short description first and loads the instructions, templates and scripts only when a task matches. A plugin packages skills, MCP connectors, commands and agents as one install that works in chat, Cowork and Claude Code.
| Piece | What it holds | Example for your team |
|---|---|---|
| Skill | Instructions, templates and scripts for one task, in a SKILL.md folder | Your proposal structure and review checklist |
| Plugin | Skills, commands, connectors and agents installed as one unit | A sales plugin with proposal and follow-up skills plus your CRM connector |
| Connector | Live access to an outside app through MCP, under each person's own login | HubSpot, Slack, Box or Microsoft 365 |
| Command | A named action you run in Cowork with /plugin-name:command | /your-plugin:proposal to draft from call notes |
Custom Claude Skills for Each Team
Each starter skill is built from your own templates and examples, so Claude's first draft follows your structure and checks. They suit Claude Cowork for small business teams and larger organizations alike.
Brand Voice and Branded Decks
A voice skill that checks drafts against your tone rules, and a deck skill that builds slides from your PowerPoint template.
Proposals and SOWs
Drafts proposals and SOWs from discovery notes in your section order, flags missing scope and leaves pricing to your team.
Reports
Turns exports and meeting notes into your weekly, monthly or client report format, with the same headings and tables every time.
Client Emails
Drafts follow-ups and status updates in your house style, using context from your connected inbox or CRM, for review before sending.
Month-End Packs
Builds the month-end pack from your close checklist, with variance commentary and a management summary ready for finance review.
Contract First Pass
Compares an incoming contract with your playbook, marks clauses that differ from your standard positions and summarizes them for review.
Want Claude to draft work the way your team does?
Book a free strategy session. Bring two or three documents your team produces every week, and we will map the skills and connectors a first plugin needs.
How We Build a Claude Cowork Plugin for Your Team
Five steps take a task from your shared drive to a plugin your whole organization can install.
1. Collect templates and examples
We gather templates, finished examples and reviewer comments, plus the rules your best people apply from memory.
2. Write the skill
Each skill gets a SKILL.md file whose description tells Claude when to use it, with the templates and scripts it needs stored beside it.
3. Test on real files
We run each skill on your real documents with Anthropic's skill-creator, the Claude skill builder that compares output with and without the skill.
4. Package as a plugin
We bundle the skills, commands and connector settings in one plugin folder with a versioned plugin.json manifest, then validate it.
5. Roll out to the org
For the Claude Cowork setup, an Owner adds the plugin and its connectors in Organization settings and sets it to available, installed by default or required.
Companies still choosing a plan start with our Claude Team and Enterprise rollout. After launch, our AI training for employees shows each team how to run its plugin and check the output.
Connectors a Cowork Plugin Can Use
Connectors give a skill live access to the apps your team works in, through the Model Context Protocol (MCP).
A plugin lists its connectors in a .mcp.json file. On Team and Enterprise an Owner adds each connector for the organization, and members sign in with their own login. Each tool can be set to Always allow, Needs approval or Blocked, so anything that writes or sends can wait for a click.
Work and chat
Slack, Microsoft 365, Notion, Asana and Jira for briefs and status updates.
Sales and support
HubSpot, Close, Intercom and Fireflies for account context and call notes.
Files and design
Box, Egnyte, Canva and Figma for source documents and brand assets.
Data and finance
Snowflake, BigQuery and Databricks for reports, plus QuickBooks, Stripe and Gusto through Anthropic's Small Business plugin.
Anthropic's own role plugins use these connectors. For an internal system without one, our AI integration services team builds a custom MCP connector, which Claude adds by its URL.
Claude Cowork Security and Governance for Skills and Plugins
Owners decide who installs what. Every third-party skill is reviewed before staff see it, and each change ships as a new version.
Who installs
Owners set each plugin to not available, available to install, installed by default or required. Enterprise adds per-group settings, and a policy switch decides whether members can upload their own skills.
Review of third-party skills
Anthropic scans every version of a directory plugin, and a person reviews each new listing. Plugins added by marketplace URL or upload sit outside that review, so we read every SKILL.md, script and .mcp.json entry before an Owner approves it.
Versioning
Each release raises the plugin.json version. Owners can compare an uploaded plugin's versions and revert in one step, and plugins synced from GitHub or GitLab update when a change reaches the default branch.
We set member publishing to Requires review, the Enterprise default since October 2, 2026, so an Owner approves each submission. In Cowork, Manual mode asks before each action, and Claude asks for explicit permission before permanently deleting files.
A Self-Hosted Option for Sensitive Files
Some files have to stay on your own servers, such as patient records, deal-room documents or HR case files. For those, we build a private assistant on a self-hosted AI stack.
Our self-hosted AI deployment runs open-weight models such as Llama, Mistral or Qwen on vLLM or Ollama inside your VPC or on your own servers. Private RAG indexes the sensitive documents there and answers with citations, with access mapped to your SSO groups.
We reuse your skills' templates, instructions and checks, so you can run Cowork plugins for everyday documents and the self-hosted assistant for restricted files.
Free Claude Skills to Try
Four free NeuralChainAI skills apply the same method to tasks many teams share.
branded-deck
Builds on-brand PowerPoint decks from your own .pptx template and brand sheet, then checks layouts and brand rules before handing over the file.
seo-audit
Runs an on-page audit of a URL or pasted HTML and returns a prioritized fix list, with a content-brief mode for new pages.
proposal-sow
Drafts proposals and SOWs from discovery notes in your section template, flags missing information and leaves price placeholders for your team.
brand-voice
Applies a fill-in brand voice guide to a draft and lists every rule the draft breaks.
All four skills are free, and each follows Anthropic's SKILL.md format, so you can read every instruction before you add it to Claude.
When you want them built on your own templates and review rules, that is the service on this page.
What Shapes the Engagement
Every Claude Cowork plugin engagement we scope starts from six factors: the teams and tasks the first plugin covers, the templates behind each task, the connectors each skill needs, any custom connector to build, your Claude plan and any files that need the self-hosted option. We quote against that scope after the strategy session, where we also answer schedule questions.
Your answers set the scope.
We confirm each factor with you before we propose anything.
Frequently Asked Questions on Claude Cowork Plugins
Plugins are available on every paid Claude plan, per Anthropic's help center, so the main vendor cost is seats plus any paid accounts your connected apps need. claude.com/pricing lists Team Standard seats at $20 per seat per month billed annually ($25 monthly) and Enterprise at $20 per seat per month billed annually with a 20-seat minimum, plus API-rate usage (checked November 3, 2026). Anthropic bills seats directly; we quote the build after the strategy session.
Start with what exists. Anthropic publishes open-source role plugins as generic starting points, including the Claude productivity plugin and plugins for sales, finance and marketing. Third-party marketplaces carry more, which we review before install. Custom skills pay off when the work runs on your own templates, playbooks and approval rules, and an Anthropic plugin can be adapted to your tools.
Cowork and plugins are available on every paid plan: Pro, Max, Ultra, Team and Enterprise. Organization controls start on Team, built for 2 to 150 people, where Owners deploy skills organization-wide and set plugin availability. A Claude Cowork Enterprise setup adds plugin settings per user group and security scanning for skills and plugins. Anthropic's Claude for Small Business page suggests Pro for one person and Team for a business with more than one.
We run sensitive work in Cowork's Manual mode, where Claude asks before each action. Connector tools that write or send stay on Needs approval, and only plugins an Owner has reviewed reach staff. Anthropic also checks plugins for malicious content when they are installed or updated, and on Enterprise the Compliance API captures Cowork sessions.
Ready to Build Claude Cowork Plugins for Your Team?
Start with a free strategy session. Share two or three documents your team produces every week and the apps they live in, and we will come back with a skill list, the connectors each one needs and a scoped proposal.