mcp-servers-essential-reference

# MCP Servers Essential Reference > Skill by [ara.so](https://ara.so) — MCP Skills collection. This skill provides expertise in selecting, installing, and configuring MCP (Model Context Protocol) servers for AI coding agents like Claude Code, Cursor, Codex, and Gemini CLI. It covers 50 essential servers across 9 categories with installation commands, security best practices, and troubleshooting. ## What MCP Servers Are The Model Context Protocol (MCP) is a standardized interface that allows AI agents to connect to external tools and services. Instead of each AI model implementing custom integrations, MCP provides a "USB-C port for AI" — one protocol that both agents and tools speak. **Key concepts:** - **Server**: A service that exposes tools/resources via MCP (e.g., GitHub API, database access) - **Transport**: How the agent connects (stdio for local, HTTP/SSE for hosted) - **Tool**: A function the agent can call (e.g., `search_github`, `run_sql_query`) - **Resource**: Read-only data the agent can access (e.g., file contents, API docs) ## The Three Critical Rules Before installing any MCP server: 1. **Don't install more than 5-7 servers** — Tool bloat degrades agent performance and reasoning quality 2. **Treat every server like untrusted code** — Pin versions, scope tokens read-only, review source 3. **Never enable writes to production** — Point at replicas, snapshots, or staging environments ## Installation Methods by Agent ### Claude Code (Desktop) ```bash # Local server (stdio transport) claude mcp add github -- npx -y @modelcontextprotocol/server-github # Hosted server (HTTP transport) claude mcp add linear --transport http # List installed servers claude mcp list # Remove a server claude mcp remove github ``` Config file: `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) ### Codex ```bash # Add server codex mcp add postgres -- npx -y @modelcontextprotocol/server-postgres # Edit config directly vim ~/.codex/config.toml ``` ### Gemini CLI Edit `~/.gemini/settings.json`: ```json { "mcpServers": { "github": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-github"], "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "${GITHUB_TOKEN}" } } } } ``` ## Manual Configuration Format All agents use similar JSON structure for `mcpServers`: ```json { "mcpServers": { "server-name": { "command": "npx", "args": ["-y", "@scope/package-name"], "env": { "API_KEY": "${ENV_VAR_NAME}" } } } } ``` **For hosted servers:** ```json { "mcpServers": { "linear": { "url": "https://mcp.linear.app", "transport": "http" } } } ``` ## Category-Based Server Selection ### Core Development (Pick 3-4) **GitHub** (Official, hosted) ```bash claude mcp add github -- npx -y @modelcontextprotocol/server-github ``` Environment: `GITHUB_PERSONAL_ACCESS_TOKEN` Capabilities: Search repos/issues/PRs, create issues, manage PRs, push commits **Filesystem** (Official, local) ```bash claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem /path/to/allowed/directory ``` ⚠️ **Security**: Only expose specific directories, never `/` or `~` **Brave Search** (Official, local) ```bash claude mcp add brave-search -- npx -y @modelcontextprotocol/server-brave-search ``` Environment: `BRAVE_API_KEY` Use case: Web search without leaving the agent context **Playwright** (Official, local) ```bash claude mcp add playwright -- npx -y @executeautomation/playwright-mcp-server ``` Capabilities: Browser automation, screenshot capture, web scraping ### Databases & Backend (Pick 1) **PostgreSQL** (Official, local) ```bash claude mcp add postgres -- npx -y @modelcontextprotocol/server-postgres postgresql://user:pass@localhost/dbname ``` ⚠️ **Critical**: Use read-only user or point at replica **Supabase** (Official, hosted) ```bash claude mcp add supabase --transport http ``` Requires: Supabase project URL and service role key **SQLite** (Community, local) ```bash claude mcp add sqlite -- npx -y @benborla/mcp-server-sqlite /path/to/database.db ``` ### Ops & Infrastructure **AWS** (Official, local) ```bash claude mcp add aws -- npx -y @modelcontextprotocol/server-aws-kb-retrieval-server ``` Environment: `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`, `AWS_REGION` **Cloudflare** (Official, hosted) ```bash claude mcp add cloudflare --transport http ``` ⚠️ **Dangerous**: Can modify DNS, workers, R2 buckets **Kubernetes** (Community, local) ```bash claude mcp add kubernetes -- npx -y @strowk/mcp-k8s-go ``` Environment: `KUBECONFIG` path ⚠️ **Production risk**: Can delete pods, services, deployments ### Productivity & Team Tools **Slack** (Community, local) ```bash claude mcp add slack -- npx -y @modelcontextprotocol/server-slack ``` Environment: `SLACK_BOT_TOKEN`, `SLACK_TEAM_ID` **Linear** (Official, hosted) ```bash claude mcp add linear --transport http ``` OAuth flow via browser **Notion** (Official, hosted) ```bash claude mcp add notion --transport http ``` **Google Drive** (Official, local - archived) ```bash claude mcp add gdrive -- npx -y @modelcontextprotocol/server-gdrive ``` Requires OAuth credentials JSON ### Fintech & Payments (⚠️ All flagged as money-moving) **Stripe** (Official, hosted/local) ```bash claude mcp add stripe -- npx -y @stripe/mcp ``` Environment: `STRIPE_SECRET_KEY` (use test keys first) **Plaid** (Official, hosted) Documentation: Check Plaid developer portal for latest MCP endpoint ⚠️ **All fintech servers**: Start in sandbox/test mode, never use production keys until you've verified agent behavior ### Web3 & Blockchain (⚠️ Irreversible transactions) **Base MCP** (Official, hosted) ```bash claude mcp add base --transport http ``` Capabilities: Deploy contracts, send transactions, query onchain data ⚠️ **Critical**: Transactions are irreversible. Use testnet first. **The Graph** (Community, local - read-only) ```bash claude mcp add thegraph -- npx -y @rainprotocol/mcp-subgraph ``` Safe for querying blockchain data via GraphQL **Dune** (Community, local - read-only) Environment: `DUNE_API_KEY` Use case: SQL queries against blockchain data ### Trading & Markets **Alpaca** (Official, local) ```bash claude mcp add alpaca -- npx -y @alpacahq/mcp-server ``` Environment: `APCA_API_KEY_ID`, `APCA_API_SECRET_KEY`, `APCA_API_BASE_URL` ⚠️ **Paper trade first**: Set `APCA_API_BASE_URL=https://paper-api.alpaca.markets` **Polygon.io** (Official, local - read-only) ```bash claude mcp add polygon -- npx -y @modelcontextprotocol/server-polygon ``` Environment: `POLYGON_API_KEY` **CoinGecko** (Official, hosted - read-only) Safe for market data queries ### Creator Tools & Media **ElevenLabs** (Official, local) ```bash claude mcp add elevenlabs -- npx -y @modelcontextprotocol/server-elevenlabs ``` Environment: `ELEVENLABS_API_KEY` **Figma** (Official, hosted/local) ```bash claude mcp add figma -- npx -y @modelcontextprotocol/server-figma ``` Environment: `FIGMA_PERSONAL_ACCESS_TOKEN` **YouTube** (Community, local) Environment: `YOUTUBE_API_KEY` Capabilities: Search videos, retrieve metadata, manage playlists ### Reasoning & Memory **Sequential Thinking** (Official, local) ```bash claude mcp add sequential-thinking -- npx -y @modelcontextprotocol/server-sequential-thinking ``` Forces step-by-step reasoning for complex problems **Memory** (Official, local) ```bash claude mcp add memory -- npx -y @modelcontextprotocol/server-memory ``` Persistent knowledge graph across sessions ## Real-World Configuration Examples ### Minimal Dev Setup (5 servers) ```json { "mcpServers": { "github": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-github"], "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "${GITHUB_TOKEN}" } }, "filesystem": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/projects"] }, "postgres": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://readonly:${DB_PASSWORD}@localhost/myapp_dev"] }, "brave-search": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-brave-search"], "env": { "BRAVE_API_KEY": "${BRAVE_API_KEY}" } }, "sequential-thinking": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-sequential-thinking"] } } } ``` ### Creator Pipeline (4 servers) ```json { "mcpServers": { "elevenlabs": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-elevenlabs"], "env": { "ELEVENLABS_API_KEY": "${ELEVENLABS_API_KEY}" } }, "figma": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-figma"], "env": { "FIGMA_PERSONAL_ACCESS_TOKEN": "${FIGMA_TOKEN}" } }, "youtube": { "command": "npx", "args": ["-y", "@benborla/mcp-server-youtube"], "env": { "YOUTUBE_API_KEY": "${YOUTUBE_API_KEY}" } }, "filesystem": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/content"] } } } ``` ### Team Productivity (5 servers) ```json { "mcpServers": { "linear": { "url": "https://mcp.linear.app", "transport": "http" }, "notion": { "url": "https://mcp.notion.so", "transport": "http" }, "slack": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-slack"], "env": { "SLACK_BOT_TOKEN": "${SLACK_BOT_TOKEN}", "SLACK_TEAM_ID": "${SLACK_TEAM_ID}" } }, "gdrive": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-gdrive"], "env": { "GOOGLE_APPLICATION_CREDENTIALS": "/path/to/credentials.json" } }, "github": { "command": "npx",