Must-Have MCP Servers in 2026 — Essential Starter Kit
The 8 MCP servers you should install before anything else. Covers every major developer workflow.
Quick Answer
The must-have MCP servers for 2026 are: Filesystem, GitHub, Fetch, PostgreSQL, SQLite, Brave Search, Memory, and Slack. Install these 8 and you'll have a complete AI development environment.
In March 2026, MCP has over 2,500 servers — but most developers need fewer than 10 to supercharge their workflow. This starter kit covers every major use case: file access, version control, databases, web search, team communication, and persistent memory.
All eight servers below work with Claude Code, Cursor, Windsurf, VS Code with Copilot, and every other MCP-compatible client.
Absolute Essentials
Filesystem MCP
Read and write your project files directly. The single most impactful MCP — unlocks true AI-assisted coding.
GitHub MCP
Manage repos, issues, and PRs from natural language. Essential if you use GitHub daily.
Fetch MCP
Fetch any URL — read docs, test APIs, scrape content. No API key required.
Database & Data
PostgreSQL MCP
Query your database in natural language. Works with any Postgres instance — local or cloud.
SQLite MCP
Zero-config database access. Perfect for local development and prototyping without a server.
Search & Knowledge
Brave Search MCP
Give your AI real-time web search. Fix the knowledge cutoff problem instantly.
Memory MCP
Persistent memory across conversations. Your AI remembers context, preferences, and project details.
Productivity Bonus
Slack MCP
Read channels, post updates, search messages. Bridge your AI with your team communication.
Complete Config for Claude Code / Claude Desktop
Copy this into your ~/.claude.json (Claude Code) or claude_desktop_config.json (Claude Desktop). Replace placeholder values with your own:
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "~/projects"]
},
"github": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-github"],
"env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_YOUR_TOKEN" }
},
"fetch": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-fetch"]
},
"postgres": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-postgres",
"postgresql://user:pass@localhost/mydb"]
},
"sqlite": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-sqlite", "~/data/mydb.sqlite"]
},
"brave-search": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-brave-search"],
"env": { "BRAVE_API_KEY": "YOUR_BRAVE_KEY" }
},
"memory": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-memory"]
},
"slack": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-slack"],
"env": { "SLACK_BOT_TOKEN": "xoxb-YOUR_TOKEN" }
}
}
}Which 3 Should I Start With?
If you're new to MCP, start with just three: Filesystem, GitHub, and Fetch. These three require minimal configuration and unlock most of what developers use AI for: editing code, managing repositories, and looking things up.
Add Brave Search once you have a free API key — it's the single best upgrade for developers who use AI for research and documentation lookup.
Frequently Asked Questions
How many MCP servers should I run at once?
Most developers run 5–10. There's no hard limit, but each server consumes a small amount of memory. Start with the essentials, then add more as you identify specific needs.
Can I use these with Cursor and Windsurf?
Yes. All eight servers in this list work with Cursor (via .cursor/mcp.json) and Windsurf (via .windsurf/mcp.json). The config format is slightly different for each tool — see our Cursor setup guide and Windsurf setup guide.
Are there free alternatives to Brave Search?
Yes — try Tavily MCP or DuckDuckGo MCP. Tavily has a generous free tier (1,000 searches/month) and excellent AI-optimized results. See our web search MCP comparison.
Install Your Starter Kit
Use our package builder to generate a ready-to-paste config for all 8 servers.