12 Best MCP Servers for Claude Desktop in 2026

{
  "mcpServers": {
    "server-name": {
      "command": "npx",
      "args": ["package-name"]
    }
  }
}

Playwright MCP Popular 28k+ stars

Built by Microsoft. Gives Claude full browser control — navigate pages, fill forms, click buttons, extract data. Uses accessibility tree snapshots instead of screenshots, making it fast and token-efficient.

• Navigate to any URL and interact with the page
• Fill forms and click buttons automatically
• Extract page content as structured data
• Take screenshots of pages or elements

// Claude Desktop config
"playwright": {
  "command": "npx",
  "args": ["@playwright/mcp@latest"]
}

GitHub

Browserbase 3.1k+ stars

Cloud-hosted browser automation. Unlike Playwright (which runs locally), Browserbase gives Claude access to real browser instances in the cloud. Useful for tasks that need persistent sessions or bypass bot detection.

• Cloud browsers that persist across sessions
• Session management with cookies and auth
• Stagehand integration for AI-native interaction

"browserbase": {
  "command": "npx",
  "args": ["@browserbasehq/mcp-server-browserbase"],
  "env": { "BROWSERBASE_API_KEY": "your-key" }
}

GitHub

Brave Search Official 700+ stars

From Brave. Lets Claude search the web in real-time. Returns web results, images, videos, and AI summaries. Free tier available (2,000 queries/month).

• Web search with real-time results
• Local search for business and location queries
• News and current events

"brave-search": {
  "command": "npx",
  "args": ["@brave/brave-search-mcp-server"],
  "env": { "BRAVE_API_KEY": "your-key" }
}

GitHub

Exa Search 3.9k+ stars

Neural search engine for developers. Better than traditional search for finding specific technical content, code examples, and documentation. Great for research tasks.

• Neural search that understands intent, not just keywords
• Content extraction from any webpage
• Configurable results with filtering options

"exa": {
  "command": "npx",
  "args": ["exa-mcp-server"],
  "env": { "EXA_API_KEY": "your-key" }
}

GitHub

DbHub 2.2k+ stars

From Bytebase. Zero-dependency database MCP server. Connect Claude to PostgreSQL, MySQL, SQLite, SQL Server, or MariaDB. Inspect schemas and run queries directly in your chat.

• Multi-database support (5 databases)
• Schema inspection — Claude sees your tables and columns
• SQL execution with results returned as tables
• Token-efficient queries

"dbhub": {
  "command": "npx",
  "args": ["@bytebase/dbhub", "--dsn", "postgres://user:pass@localhost:5432/mydb"]
}

GitHub

Filesystem Official

From the official MCP servers repo. Gives Claude safe, controlled access to your local files. Read, write, search, and manage files with configurable access controls.

• Read/write files with permission controls
• Directory listing and file search
• Move, copy, delete with safety checks

"filesystem": {
  "command": "npx",
  "args": ["@modelcontextprotocol/server-filesystem", "/path/to/allowed/dir"]
}

GitHub

AWS MCP Servers Official 8.3k+ stars

From Amazon Web Services. A suite of 20+ specialized MCP servers for different AWS services. Manage EC2, S3, Lambda, DynamoDB, CloudFormation, and more directly from Claude.

• AWS Documentation — search and reference AWS docs
• CDK — generate and modify infrastructure as code
• Bedrock — manage AI/ML models on AWS
• Cost Explorer — analyze spending and optimize

"aws-docs": {
  "command": "npx",
  "args": ["@aws/aws-documentation-mcp-server"]
}

GitHub

Cloudflare Official 3.4k+ stars

From Cloudflare. Deploy and manage Workers, KV stores, R2 buckets, and D1 databases. Great for building and deploying edge applications directly from Claude.

• Workers — deploy serverless functions
• KV & R2 — manage key-value and object storage
• D1 — run SQL queries on edge databases

"cloudflare": {
  "command": "npx",
  "args": ["@cloudflare/mcp-server-cloudflare"]
}

GitHub

Notion Official 3.9k+ stars

From Notion. Query databases, create pages, manage content in your Notion workspace. Useful for project management, note-taking, and knowledge base workflows.

• Search across all your Notion pages and databases
• Create pages with rich content and properties
• Query databases with filters and sorting

"notion": {
  "command": "npx",
  "args": ["@notionhq/notion-mcp-server"],
  "env": { "NOTION_API_KEY": "ntn_..." }
}

GitHub

Memory Official

From the official MCP servers repo. Gives Claude persistent memory across conversations using a knowledge graph. Store facts, relationships, and context that survives between chats.

• Knowledge graph storage for entities and relations
• Persistent across Claude Desktop sessions
• Search and retrieve stored knowledge

"memory": {
  "command": "npx",
  "args": ["@modelcontextprotocol/server-memory"]
}

GitHub

DeFi MCP 12 Tools

A comprehensive MCP server for DeFi and cryptocurrency data. Get real-time token prices, check wallet balances across 9 blockchains, estimate gas fees, and get DEX swap quotes — all from Claude Desktop.

• Token prices — real-time prices for any token (CoinGecko)
• Wallet balances — Ethereum, Solana, Base, Arbitrum, Polygon, BSC, Optimism, Avalanche, Fantom
• Gas estimation — current gas prices and fee estimates
• DEX quotes — swap quotes from Jupiter (Solana) and Uniswap (EVM)
• Token metadata — name, symbol, decimals, contract info
• Protocol data — TVL and DeFi stats via DeFi Llama

Example prompts:

• "What's the current price of ETH and SOL?"
• "Check the balance of vitalik.eth on Ethereum"
• "Get a swap quote for 1 SOL to USDC on Jupiter"
• "What's the current gas price on Ethereum?"
• "Show me the top DeFi protocols by TVL"

// Claude Desktop config
"defi-mcp": {
  "command": "npx",
  "args": ["defi-mcp"]
}

// Or run the hosted REST API (no install needed):
curl https://agent-gateway-kappa.vercel.app/v1/defi-mcp/api/prices?ids=bitcoin,ethereum,solana

GitHub · Setup Guide · API Catalog

Firecrawl 5.6k+ stars

Turn any website into clean, structured data. Firecrawl handles JavaScript rendering, bot detection, and pagination automatically. Perfect for research, competitor analysis, and data collection.

• Scrape any URL and get clean markdown
• Crawl entire sites with configurable depth
• Extract structured data from web pages
• Search the web and get full content

"firecrawl": {
  "command": "npx",
  "args": ["-y", "firecrawl-mcp"],
  "env": { "FIRECRAWL_API_KEY": "your-key" }
}

GitHub