Claude Code Team Collaboration Guide: Shared CLAUDE.md, MCP Config & Auto Code Review

Solo, Claude Code is a productivity beast. Two people? Starts to get messy. Three or more—everyone has their own CLAUDE.md, MCP configs are a jumble, coding styles clash, and bugs get reverted across branches. This isn't Claude Code's fault—it's the lack of team discipline. Here's a battle-tested playbook to turn Claude Code from a personal tool into team infrastructure.

## 1. Centralize CLAUDE.md: One Source of Truth CLAUDE.md is your project's behavior guide for Claude Code. **Rule #1: CLAUDE.md lives in the project root, version-controlled, no one modifies it solo.** ### 1.1 Standard Template A team-ready CLAUDE.md should include: ```markdown # Project Conventions ## Stack - Frontend: React 18 + TypeScript + Tailwind - Backend: Python 3.12 + FastAPI - Database: PostgreSQL 16 - Package Manager: pnpm (frontend) / uv (backend) ## Code Style - TypeScript: strict mode - Python: ruff formatter, line-length=100 - All API returns must include type definitions - Components: PascalCase, utilities: camelCase ## Testing - Every new feature needs unit tests - Minimum 80% coverage - Test files go in `__tests__/` ## Commit Convention - feat: new feature - fix: bug fix - refactor: code change - docs: documentation - test: testing ``` ### 1.2 Git Hook Protection Commit CLAUDE.md to Git and add a pre-commit hook: ```bash # .git/hooks/pre-commit if git diff --cached --name-only | grep -q "CLAUDE.md"; then echo "⚠️ CLAUDE.md modified—discuss changes with the team first" exit 1 fi ``` Now everyone's Claude Code behaves the same way. No more "A tells Claude to use tabs while B tells it to use spaces." ## 2. Share MCP Config: Project-Level > Personal MCP (Model Context Protocol) extends Claude Code's capabilities. The team pitfall: **each member installs different MCP servers, so Claude generates code depending on different external capabilities.** ### 2.1 Project-Level MCP Config Claude Code supports project-level `.mcp.json`: ```json { "mcpServers": { "playwright": { "command": "npx", "args": ["@playwright/mcp"] }, "git": { "command": "npx", "args": ["@anthropic-ai/mcp-git"] } } } ``` Check this into Git. Now everyone's Claude has the exact same toolset. ### 2.2 Declare MCP Requirements in CLAUDE.md ```markdown ## MCP Config - Default MCP servers: playwright (browser testing), git (version queries) - Adding a new MCP server requires team discussion - Third-party MCP servers must be documented with source and purpose ``` ### 2.3 Recommended Team MCP Servers | MCP Server | Purpose | Pick | |:----------|:--------|:----:| | `@playwright/mcp` | Browser testing | ⭐ | | `@anthropic-ai/mcp-git` | Git operations | ⭐ | | `@modelcontextprotocol/mcp-sqlite` | Database queries | | | `@anthropic-ai/mcp-filesystem` | File operations | ⭐ | | `@modelcontextprotocol/mcp-github` | GitHub API | | ## 3. Automate Code Review Claude Code's most underrated team feature: **automated code review.** ### 3.1 Local Review Command Before pushing a PR: ```bash claude -p "Review the diff between current branch and main. Check for: 1) CLAUDE.md violations 2) potential bugs 3) missing type definitions 4) outdated comments" ``` Wrap it in a Makefile: ```makefile review: git diff main...HEAD | claude -p "Review against @CLAUDE.md rules" ``` ### 3.2 CI Auto-Review with GitHub Actions ```yaml name: Claude Code Review on: [pull_request] jobs: review: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - name: Claude Review env: ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }} run: | git diff origin/main...HEAD | \ npx @anthropic-ai/claude-code review - \ --context CLAUDE.md \ --output review.md - uses: actions/upload-artifact@v4 with: name: claude-review path: review.md ``` ### 3.3 Review Checklist in CLAUDE.md ```markdown ## Review Checklist ### Security - [ ] No hardcoded API keys - [ ] Parameterized SQL queries - [ ] Input validation - [ ] No known vulnerable dependencies ### Quality - [ ] Unit tests included - [ ] Error handling complete - [ ] Adequate logging - [ ] No magic numbers ### Performance - [ ] No unnecessary I/O in loops - [ ] Caching considered - [ ] No N+1 database queries ``` ## 4. Real Team Workflow A 3-person team's typical Claude Code day: | Time | Member A | Member B | Member C | |:----|:---------|:---------|:---------| | 09:00 | `claude "look at issue #42, add export feature"` | `claude "review PR #40"` | `claude "write unit tests"` | | 10:00 | Claude generates code, git commit | Submits review | Tests pass, 85% coverage | | 11:00 | Push PR, CI auto-reviews | Starts new feature | Reviews PR #41 | | 14:00 | Handles review feedback | `claude "fix review issues"` | Merges PR #40 | Key takeaways: - **Unified CLAUDE.md** → consistent code style across the team - **Shared MCP config** → identical Claude behavior everywhere - **Automated review** → human reviewers focus on business logic, not formatting ## 5. Caveats 1. **CLAUDE.md evolves** — update it as the project grows, but discuss changes first 2. **MCP server security** — don't install unvetted third-party MCP servers 3. **AI review ≠ final review** — catches style issues, but business correctness needs a human 4. **Introduce gradually** — start with unified CLAUDE.md for a week, add auto-review later 5. **Lowered communication overhead, not eliminated** — Claude handles formatting, humans debate design --- *More from the Claude Code series:* - [Hands-on: Build a Web App in 10 Minutes with AI](/post/claude-code-todo-project-2026) - [Advanced: CLAUDE.md & MCP Configuration](/post/claude-code-advanced-2026) - [Pitfalls: 10 Mistakes Beginners Make](/post/claude-code-pitfalls-2026) - [2026 AI Coding Tools Buyers Guide](/post/ai-coding-tools-guide-2026) - [Can Claude Code Be Used in Hong Kong?](/post/claude-code-xianggang-2026)