Recommended MCP Servers for Claude Code

MCP (Model Context Protocol) servers extend Claude Code with external capabilities — databases, documentation, file systems, and custom tools.

Recommended Servers

Context7

Purpose: Live library documentation lookup. Fetches current docs for any library, bypassing Claude’s knowledge cutoff.

claude mcp add --scope user --transport stdio context7 -- npx -y @context7/mcp@latest

When to use: When working with rapidly-evolving libraries (LangChain, Next.js, etc.) where API changes frequently.

PostgreSQL MCP

Purpose: Direct database introspection — schema browsing, query execution, migration validation.

claude mcp add --scope project --transport stdio postgres -- \
  npx -y @anthropic/mcp-postgres "$DATABASE_URL"

When to use: Projects with PostgreSQL databases. Enables Claude to inspect schemas, validate migrations, and write accurate queries.

Filesystem MCP

Purpose: Scoped file access outside the working directory — read config files, logs, or shared resources.

claude mcp add --scope project --transport stdio filesystem -- \
  npx -y @anthropic/mcp-filesystem /path/to/allowed/directory

When to use: When Claude needs access to files outside the project root (e.g., shared config, monorepo siblings).

Playwright (Browser Automation)

Purpose: Browser automation for debugging, visual testing, and e2e validation in web projects.

Recommended: Playwright CLI (token-efficient, designed for coding agents with shell access):

# Install (or use setup.sh --playwright)
npm install -g @playwright/cli@latest
playwright-cli install --skills

# Usage
playwright-cli open http://localhost:<port>
playwright-cli click "Login button"
playwright-cli screenshot

Alternative: Playwright MCP (for Docker/CI environments without shell access):

claude mcp add --scope project --transport stdio playwright -- \
  npx -y @playwright/mcp@latest --headless

Or via Docker:

claude mcp add --scope project --transport stdio playwright -- \
  docker run -i --rm --init mcr.microsoft.com/playwright/mcp

When to use: Web projects (web-static, web-dynamic, any project with a browser UI). Use CLI for local development, MCP for containerized environments.

When to Use MCP vs Built-in Tools

Capability	Built-in Tool	MCP Server	Prefer
Read project files	Read, Glob, Grep	Filesystem MCP	Built-in (already scoped)
Search the web	WebSearch, WebFetch	—	Built-in
Library docs lookup	WebFetch (manual URL)	Context7	MCP (structured, current)
Database schema	Bash(psql …)	PostgreSQL MCP	MCP (safer, structured)
Session continuity	CLAUDE.md, specs, phase recaps	—	Built-in
Files outside project	Bash(cat …)	Filesystem MCP	MCP (scoped, auditable)
Browser automation	Bash(npx playwright test)	Playwright CLI / MCP	CLI (token-efficient)

Scope Guidance

Scope	Meaning	Use For
--scope user	Available in all projects	Context7 (general-purpose tools)
--scope project	Only available in this project	PostgreSQL MCP, Filesystem MCP, Playwright MCP (project-specific)

Managing MCP Servers

claude mcp list                    # list installed servers
claude mcp remove <name>           # remove a server
/mcp                               # manage from inside a session

Tips

• Keep the number of MCP servers small — each adds startup latency and context overhead.
• Use --scope project for databases and sensitive resources to avoid leaking access across projects.
• Test MCP servers in a fresh session after installing to verify they connect.
• If an MCP server is slow to start, check that the underlying service (database, API) is running.