Best MCP Servers for Developers in 2026: Web Data, Code, Files, and More

Contains affiliate links. Disclosure.

The Model Context Protocol (MCP) is an open standard (originated by Anthropic, now community-governed) for connecting AI assistants like Claude (Desktop and Code), ChatGPT, VS Code, and Cursor to external tools. The ecosystem has grown quickly, with thousands of community-built servers now available. This guide covers the ones that are most useful for developers, with setup instructions and honest notes on when each fits.

How MCP Servers Work

MCP servers are local processes (Node.js or Python) that expose tools as JSON-RPC methods. Claude Desktop loads them on startup via claude_desktop_config.json:

{
  "mcpServers": {
    "server-name": {
      "command": "npx",
      "args": ["-y", "@package/mcp-server"],
      "env": { "API_KEY": "your_key" }
    }
  }
}

---

Web Data MCP Servers

1. Apify MCP Server

Access 30,000+ web scraping Actors from Claude: Google Maps, LinkedIn, Amazon, Instagram, YouTube, and more.

The simplest path is the hosted server at https://mcp.apify.com/?fpr=use-apify, which uses OAuth (no token to paste into a config file) and is the option Apify recommends. Add it as a remote connector in Claude Desktop, or point any MCP host at the URL.

If you prefer a local server, run it via npx and scope it to one or two Actors with --actors so the model only sees the tools you need:

{
  "mcpServers": {
    "apify": {
      "command": "npx",
      "args": ["-y", "@apify/actors-mcp-server", "--actors", "apify/website-content-crawler"],
      "env": { "APIFY_TOKEN": "your_token" }
    }
  }
}

Tools exposed: Run Apify Actors, read dataset items, search and list available Actors.

Best for: Structured data extraction from named platforms (Amazon prices, Maps leads, LinkedIn profiles).

New to Claude and want to test this stack before paying for anything? You can try Claude free for a week and pair it with Apify's $5 monthly free credits.

Setup guide → | Get Apify token →

---

2. Firecrawl MCP Server

Convert any URL to Markdown for Claude context.

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

Tools exposed: scrape (URL → Markdown), crawl (full site crawl), map (URL discovery).

Best for: RAG pipelines, reading documentation, competitive research.

Setup guide → | Get Firecrawl API key →

---

3. Bright Data MCP Server

Browser-level web scraping with anti-bot bypass.

{
  "mcpServers": {
    "brightdata": {
      "command": "npx",
      "args": ["-y", "@brightdata/mcp"],
      "env": {
        "API_TOKEN": "your_token",
        "WEB_UNLOCKER_ZONE": "your_zone"
      }
    }
  }
}

Tools exposed: scrape_as_markdown, scrape_as_html, web search.

Best for: Sites protected by Cloudflare, CAPTCHA, or heavy anti-bot.

Setup guide → | Get Bright Data access →

---

Developer Productivity MCP Servers

4. GitHub MCP Server

Read repositories, issues, PRs, and file contents from Claude. GitHub now maintains its own official MCP server (with a hosted remote endpoint plus a self-hosted option), which is the path to prefer for new setups. Check GitHub's MCP documentation for the current install command and authentication flow, since it has moved since the earliest community packages.

Best for: Code review assistance, PR summarization, issue triage.

---

5. Filesystem MCP Server (Official)

Read, write, and search files in allowed directories.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/projects"]
    }
  }
}

Best for: Local code editing, document processing, file organization.

---

6. PostgreSQL MCP Server (Official)

Query your database directly from Claude.

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://localhost/mydb"]
    }
  }
}

Best for: Data analysis, SQL generation from natural language, schema exploration.

---

7. Slack MCP Server (Official)

Read channels, post messages, and search Slack history.

{
  "mcpServers": {
    "slack": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-slack"],
      "env": {
        "SLACK_BOT_TOKEN": "xoxb-...",
        "SLACK_TEAM_ID": "T..."
      }
    }
  }
}

Best for: Summarizing conversations, drafting updates, searching team discussions.

---

8. Puppeteer MCP Server

Browser automation from Claude: click, type, screenshot.

{
  "mcpServers": {
    "puppeteer": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-puppeteer"]
    }
  }
}

Best for: Manual site testing, scraping with Claude-guided interaction, form automation. Microsoft's Playwright MCP server is a well-maintained alternative in the same category and worth checking if you want active upstream support.

---

Comparison Table

MCP Server	Category	Tools	Auth Required	Free Tier
Apify	Web data	30,000+ Actors	API token or OAuth	$5/month credits
Firecrawl	Web data	scrape, crawl, map	API key	500 credits
Bright Data	Web data	scrape, search	API token + zone	Trial available
GitHub	Code	read, write, search	PAT	Yes
Filesystem	Files	read, write, search	None (local paths)	Yes
PostgreSQL	Database	query, list tables	Connection string	Yes
Slack	Comms	read, post, search	Bot token	Yes
Puppeteer	Browser	navigate, click, type	None	Yes

Pricing verified June 2026. Plans change frequently — check the vendor’s official pricing page for current rates.

---

Setup Best Practices

1. Limit active servers to 5 or 6. Each one starts a subprocess and adds tool definitions to the context.
2. Restart Claude Desktop after any config change.
3. Use env vars for secrets. Never hardcode API keys in config, and prefer OAuth-based hosted servers (like Apify's) where available.
4. Test each server with a simple Claude prompt before building complex workflows
5. Check logs at ~/Library/Logs/Claude/mcp-*.log when a server fails to load

---

FAQ

Frequently Asked Questions

Start with the MCP Filesystem server (for reading and writing files), the GitHub server (for code search and issue management), and one web data server: Apify MCP if you need structured extraction from many sites, or Firecrawl MCP if you primarily want clean Markdown from URLs. These cover the large majority of developer use cases and are free or low-cost to start.

Use Apify MCP when you need structured extraction from specific websites (Google Maps, Amazon, LinkedIn, social media). Use Firecrawl MCP when you need fast URL-to-Markdown conversion for RAG or LLM consumption. Use Bright Data MCP when targets use heavy anti-bot protection (Cloudflare enterprise, major e-commerce). Jina Reader MCP is good for simple public documentation.

The MCP protocol itself is production-ready. Anthropic uses it in Claude Desktop and Claude Code, and editors like Cursor and VS Code support it too. Individual server quality varies. Reference servers (filesystem, Postgres, and similar) and well-maintained vendor servers (Apify, Firecrawl, GitHub's official server) are reliable. Community servers may have gaps. Always test with your specific workflow before using in production.

Add the Apify or Firecrawl MCP server (the hosted Apify endpoint via OAuth is the quickest option) and restart Claude Desktop. Then simply ask Claude: 'Use the Apify tools to scrape all coffee shop listings from Google Maps in Austin, TX and return them as a table.' Claude will call the appropriate Actor automatically. For complex multi-step research, enable extended thinking for better planning.