The Definitive Guide to Claude Code for Teams

Your team is already using Claude Code. Some on personal Pro accounts, some burning through API credits, some sharing logins they shouldn't be sharing. Whether you're an engineering team, sales org, or ops group - here's how to do it properly.

The Plan Landscape

Pricing as of January 2026 - verify at claude.ai/pricing before deciding.

Important caveat: Max plans ($100 and $200) are only available for individual accounts, not team accounts. If you want the higher token limits of Max, each team member needs their own individual Max subscription and can expense/get reimbursed by the company.

API pricing (if you're tracking costs directly):

• Opus 4.5: $5 input / $25 output per 1M tokens
• Sonnet 4.5: $3 input / $15 output per 1M tokens
• Haiku: $0.80 input / $4 output per 1M tokens

A typical Claude Code session runs $0.50-5 depending on complexity. Heavy users spend $20-50/day.

When to Use What

Individual (1 person)

Start with Pro ($20/mo). It's enough to learn the platform and handle most tasks. When you start hitting limits regularly, upgrade to Max $100.

Small team (2-4 people)

Skip the Team plan. Put everyone on Max $100 or Max $200 individually and have them expense it. Max plans are only available for individual accounts (not team plans), so reimbursement is the way to go. You don't get admin controls, but you also don't pay the premium seat fees or deal with pooled usage complexity.

Total cost: $200-800/mo for 2-4 people with high limits each.

Larger team (5+ people)

Team plan starts making sense here. You get:

• Admin controls and spending limits per user
• Pooled usage across the team
• Shared billing and usage analytics
• Role-based access

The catch: premium seats for Claude Code access cost extra (check current pricing), and you pay API rates for overage. Budget unpredictability is real.

Enterprise

If you need any of these, you need Enterprise:

• SSO and SCIM provisioning
• Audit logs and compliance API
• Custom data retention
• Dedicated account management

Contact Anthropic sales for pricing. It's negotiated per-org.

The Cost Reality

Claude Code billing is usage-based, not flat-rate. This creates budget unpredictability.

How to track:

• Type /cost during any session to see current spend
• Use tools like ccusage to track usage across sessions
• Set spending limits per user in Team/Enterprise admin settings

What burns tokens fast:

• Large file sets (every file Claude reads costs tokens)
• Long conversations (context accumulates)
• Searching across many files
• Running multiple parallel sessions

What saves tokens:

• Good CLAUDE.md files (Claude doesn't have to explore)
• Specific prompts (less back-and-forth)
• Using /compact to summarize long conversations
• Closing sessions when done instead of leaving them open

Two Paths: Web vs CLI

Claude on the Web (claude.ai) - Best for non-technical teams or getting started. Use Projects to organize work, upload files, set custom instructions, and share conversations. No installation required.

Claude Code (CLI) - Best for technical teams. Full file system access, runs commands, integrates with your dev tools. More powerful, steeper learning curve.

Most teams use both. Sales and ops use the web. Engineering uses the CLI. Both can share knowledge through documentation.

Setting Up Your Team (Web)

For teams using claude.ai Projects:

Shared Projects

Create a project for each major workstream. Add:

• Files - Upload documents, data, reference material (up to 200K tokens / ~500 pages)
• Custom instructions - Team voice, formatting rules, process guidelines
• Shared conversations - Let teammates see useful chats in the activity feed

Team members can have "view" (use the project) or "edit" (modify instructions and files) access.

Example project setup for a sales team:

• Project: "Outbound Sales"
• Files: Product one-pager, competitor comparison, pricing guide, case studies
• Instructions: "Write in a conversational but professional tone. Keep emails under 150 words. Always include a specific reason for reaching out."

Now every conversation in that project follows your team's playbook.

Setting Up Your Team (CLI)

For teams using Claude Code:

1. Shared CLAUDE.md

Create a CLAUDE.md file in your shared workspace. This is the single most important team asset - it teaches Claude your team's context and preferences.

For engineering teams (check into git):

# Project Name
 
## Overview
[What this project does, who it's for]
 
## Tech Stack
- Frontend: React, TypeScript
- Backend: Python, FastAPI
 
## Conventions
- Use pnpm, not npm
- Run `pnpm test` before committing
 
## Things Claude Gets Wrong
- Don't modify /auth directly - use the auth service
- Always use UTC for timestamps

For sales/ops teams (store in shared folder):

# Sales Team - Claude Instructions
 
## Our Process
- We use HubSpot for CRM
- Outreach sequences live in /campaigns/
- Prospect research goes in /prospects/[company]/
 
## Writing Style
- Keep emails under 150 words
- Always personalize the first line
- Never use "I hope this email finds you well"
 
## Data Sources
- Company info: Pull from LinkedIn, Crunchbase
- Contact info: Use Apollo or ZoomInfo exports
- Financials: Check latest 10-K if public
 
## Things Claude Gets Wrong
- Don't assume job titles - verify on LinkedIn
- Always check if we've contacted this company before

The pattern that works: Every time Claude makes a mistake, add it to CLAUDE.md. "Every mistake becomes a rule." Anthropic's internal team contributes to their shared CLAUDE.md multiple times per week.

2. Project settings (.claude/settings.json)

For engineering teams: Check this into git to share permission defaults:

{
  "permissions": {
    "allow": [
      "Bash(npm test:*)",
      "Bash(git status)",
      "Bash(git diff:*)"
    ],
    "deny": [
      "Bash(rm -rf:*)",
      "Bash(git push --force:*)"
    ]
  }
}

This prevents Claude from running destructive commands while allowing common safe operations without prompts. Non-technical teams can skip this.

3. Shared slash commands (.claude/commands/)

Create team workflows as markdown files. Anyone can type /command-name to run them.

For engineering - /review:

Review the changes in this PR for:
1. Security issues
2. Performance problems
3. Code style violations
4. Missing tests
 
Be specific about line numbers and suggest fixes.

For sales - /research:

Research this company for outreach:
1. What they do (one sentence)
2. Recent news or funding
3. Key decision makers with titles
4. Potential pain points we solve
5. Personalization hooks for email
 
Save to /prospects/[company-name]/research.md

Commands that pay off:

4. GitHub Actions integration (engineering teams)

Automate PR reviews with the official Claude Code GitHub Action:

name: Claude Code Review
on:
  pull_request:
    types: [opened, synchronize]
  issue_comment:
    types: [created]
 
jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: anthropics/claude-code-action@beta
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}

Claude reviews every PR automatically. Team members can @mention Claude in PR comments for follow-up questions. See the action docs for configuration options.

Case Studies: What Teams Actually Do

From Anthropic's internal teams:

incident.io uses their CLAUDE.md as onboarding documentation for both Claude and new hires. They built custom slash commands that let anyone on the team add discoveries to the shared knowledge base.

If You're a Team Member: How to Pitch This

Want Claude Code for your team but need buy-in? Here's a message you can copy and send to your manager:

Hey [Manager],

I've been reading about Claude Code - it's Anthropic's AI assistant that runs in the terminal. Unlike ChatGPT, it can actually read files, run commands, and make changes directly.

Some things teams are doing with it:
- 3x faster debugging (Anthropic's security team)
- Automated prospect research (sales teams cutting research time by 80%)
- Shared team knowledge that persists across sessions

The pricing:
- Individual: $20-200/mo per person depending on usage
- Team plan (5+ people): $30/user/mo + usage, with admin controls
- Enterprise: Custom pricing with SSO/audit logs

I think it could speed up [specific task you do] significantly. Would you be open to a small pilot? We could start with 2-3 people on individual Pro accounts ($20/mo each) and see if the productivity gains justify rolling it out wider.

Here's a good overview: [link to this post when you send it]

Let me know what you think.

If You're a Manager: How to Roll This Out

Step 1: Get your own account first (5 minutes)

1. Go to claude.ai and create an account
2. Subscribe to Pro ($20/mo) to try it yourself
3. Install Claude Code: npm install -g @anthropic-ai/claude-code
4. Run claude in a project folder and ask it to explain what's there

Don't roll out what you haven't used yourself.

Step 2: Decide on the plan

Step 3: Pilot with 2-3 people (1-2 weeks)

Pick your most adventurous team members. Give them accounts. Ask them to:

• Use it for real work, not toy projects
• Track what works and what doesn't
• Note any security or workflow concerns

Step 4: Set up team infrastructure (1 hour)

Before wider rollout:

1. Create a CLAUDE.md in your shared workspace (see templates above)
2. For engineering: add .claude/settings.json with permission defaults
3. Create 1-2 slash commands for common workflows

Step 5: Announce to the team

Team,

We're rolling out Claude Code - an AI assistant that can read files, run commands, and make changes directly.

**To get started:**
1. You should have received an invite to our [Team/Pro] plan
2. Install it: `npm install -g @anthropic-ai/claude-code`
3. Run `claude` in any project folder and start asking questions

**What it can do:**
- Explain files and projects ("what does this do?")
- Research and analyze ("summarize these documents")
- Make changes ("update this spreadsheet", "draft an email")
- Run our team workflows (type `/` to see available commands)

**Important:**
- We have a shared CLAUDE.md with team conventions - Claude reads it automatically
- Type `/cost` to see your usage
- Don't paste sensitive credentials into prompts

**Resources:**
- Claude Code for Teams: [share this post with your team]
- Claude Code for Engineers: punnkam.com/writing/claude-code-engineers
- Claude Code for Everyone: punnkam.com/writing/claude-code-everyone

Questions? Ping me directly.

Step 6: Iterate on CLAUDE.md

After the first week:

• Collect feedback on what Claude gets wrong
• Add those mistakes to CLAUDE.md
• Share wins in Slack to drive adoption

Getting Started Checklist

• Decide on plan (Pro/Max for small teams, Team for 5+, Enterprise for compliance)
• Create CLAUDE.md in shared workspace with team context and conventions
• (Engineering) Add .claude/settings.json with permission defaults
• Create 2-3 slash commands for common workflows
• (Engineering) Set up GitHub Action for automated PR review
• Hold a team session to share workflows and calibrate prompts
• Assign a "CLAUDE.md maintainer" to keep it updated

The real unlock: Individual Claude Code makes one person faster. Team Claude Code creates institutional memory that compounds - every lesson learned, every mistake avoided, every workflow optimized gets encoded and shared. The AI gets smarter about your team's work over time, not because the model improves, but because your documentation does.

Sources:

• Claude Pricing
• Claude Code Pricing Guide
• Enterprise Claude Code Guide
• How Anthropic Teams Use Claude Code