10 MCP Servers Every Developer Should Set Up (With Step-by-Step Guide)

Cole Bennett — Sat, 31 Jan 2026 23:56:22 +0000

The Model Context Protocol (MCP) ecosystem has exploded. When Anthropic launched MCP in November 2024, there were roughly 100 servers. Today, the PulseMCP directory lists over 7,800 -- and major players like OpenAI, Google, and Microsoft have all adopted the standard.

Here's the thing -- most of those 7,800 servers are niche, experimental, or already unmaintained. After testing dozens of them, these are the 10 that actually matter for day-to-day development work.

If you're new to MCP: It's a standard protocol that lets AI assistants (Claude, ChatGPT, Cursor, etc.) connect to external tools and data sources. Think of it as USB-C for AI -- one standard interface, many devices.

Prerequisites

Before configuring any MCP servers, make sure you have:

Node.js 18+ (required for most servers)
Python 3.10+ (for Python-based servers like Git and Fetch)
Docker (optional, but recommended for database servers)

Config file locations:

Claude Desktop (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json
Claude Code: ~/.claude.json or use claude mcp add
Cursor: ~/.cursor/mcp.json
VS Code: Command Palette > "MCP: Open User Configuration"

1. Filesystem Server

What it does: Read, write, search, and manage files on your local system.

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

Without filesystem access, your AI can only respond to content you paste in. With it, the AI explores your entire codebase, reads docs, and writes files directly.

Security note: The path argument restricts access to specific directories -- don't point it at your home directory.

2. GitHub Server

What it does: Manage repositories, issues, pull requests, and search code via the GitHub API.

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here"
      }
    }
  }
}

GitHub now maintains the official server themselves. It enables AI to create PRs, review code, manage issues, and search across repositories.

3. Git Server

What it does: Local Git operations -- clone, commit, branch, diff, log. Works with any Git repo, not just GitHub.

{
  "mcpServers": {
    "git": {
      "command": "uvx",
      "args": ["mcp-server-git", "--repository", "/path/to/git/repo"]
    }
  }
}

This complements the GitHub server by handling local operations. Your AI can commit changes, create branches, review diffs, and manage worktrees independently.

4. PostgreSQL Server

What it does: Read-only access to PostgreSQL databases for schema inspection and queries.

{
  "mcpServers": {
    "postgres": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "POSTGRES_URL", "mcp/postgres"],
      "env": {
        "POSTGRES_URL": "postgresql://user:pass@host:5432/mydb"
      }
    }
  }
}

Critical: Use a read-only database user. Never connect production databases directly. Docker is recommended to keep your database credentials isolated.

5. Memory Server (Knowledge Graph)

What it does: Persistent memory through a local knowledge graph. Entities, relations, and observations survive between sessions.

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"],
      "env": {
        "MEMORY_FILE_PATH": "/path/to/memory.jsonl"
      }
    }
  }
}

AI assistants typically forget everything between sessions. The memory server stores facts, relationships, and context in a queryable knowledge graph -- so your AI remembers project decisions and ongoing work.

6. Playwright Server (Browser Automation)

What it does: Browser automation using the accessibility tree. Navigate, click, fill forms, take screenshots.

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["-y", "@playwright/mcp@latest"]
    }
  }
}

Microsoft's official MCP server, ranked #2 globally. Uses the accessibility tree for deterministic interactions -- faster, lighter, and compatible with Chromium, Firefox, and WebKit.

7. Slack Server

What it does: Read messages, post updates, search channels, and interact with Slack workspaces.

{
  "mcpServers": {
    "slack": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-slack"],
      "env": {
        "SLACK_BOT_TOKEN": "xoxb-your-token",
        "SLACK_TEAM_ID": "T0123456789"
      }
    }
  }
}

Setup note: You'll need to create a Slack app first with the appropriate OAuth scopes (channels:history, channels:read, chat:write, users:read at minimum).

8. Notion Server

What it does: Read and write to Notion workspaces -- databases, pages, blocks.

{
  "mcpServers": {
    "notion": {
      "command": "npx",
      "args": ["-y", "@notionhq/notion-mcp-server"],
      "env": {
        "NOTION_TOKEN": "ntn_****"
      }
    }
  }
}

Connects your documentation, project management, or knowledge base living in Notion directly to your AI workflow. Notion also offers a hosted version with OAuth if you prefer not to manage API keys.

9. Brave Search Server

What it does: Web search with a privacy focus. Returns search results, local business data, images, videos, and news.

{
  "mcpServers": {
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-brave-search"],
      "env": {
        "BRAVE_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

The free tier includes 2,000 queries per month. Sign up at brave.com/search/api.

10. Fetch Server

What it does: Fetch web content and convert HTML to markdown for LLM consumption.

{
  "mcpServers": {
    "fetch": {
      "command": "uvx",
      "args": ["mcp-server-fetch"]
    }
  }
}

Simple but essential. Handles webpage content conversion to AI-friendly text. Respects robots.txt by default.

Verification

After configuring your servers, restart your AI client. In Claude Desktop, look for the hammer icon in the chat interface -- if it's there, your MCP servers are connected.

Troubleshooting Common Issues

PATH problems on macOS: Apps launched from Finder don't inherit your shell PATH. Use absolute paths like /opt/homebrew/bin/npx instead of just npx.

JSON syntax errors: A missing comma will break your entire config silently. Use a JSON validator.

Debug logs: Check ~/Library/Logs/Claude/mcp-server-*.log on macOS for detailed error output.

MCP Inspector: Run npx @modelcontextprotocol/inspector to test individual servers in isolation.

Where to Find More Servers

PulseMCP Directory -- 7,800+ servers, searchable
Smithery -- 2,200+ servers
Official MCP Repository on GitHub

If you want to build your own MCP server, check out this step-by-step tutorial on building MCP servers -- it walks through the full process from scratch.

And if you're using Claude Code specifically, this guide covers the complete setup including MCP configuration.

This article was originally published on PulseMark.ai, where we cover AI development tools, model analysis, and practical engineering tutorials daily.

DEV Community: Cole Bennett