20 Best MCP Servers 2026 — The Definitive Setup Guide for Claude & Cursor

If you’re using Claude, Cursor, or any modern AI coding assistant and haven’t set up MCP servers yet, you’re leaving most of the capability on the table.

Model Context Protocol (MCP) is Anthropic’s open standard that lets AI tools connect to external data sources — your database, your GitHub repos, your browser, your Slack — without you having to paste anything into a chat window. Set it up once, and your AI assistant can actually do things instead of just telling you what to do.

We tested 20+ MCP servers across databases, developer tools, productivity apps, and automation platforms. Here’s what’s worth installing.

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

Quick Comparison: Top 20 MCP Servers

The Must-Install MCP Servers

1. filesystem — Read & Write Local Files

Source: Official Anthropic | Install: npx @modelcontextprotocol/server-filesystem

The foundation. filesystem gives Claude or Cursor the ability to read any file on your machine, create new files, move them, and search through directories. Without this, your AI only sees what you paste.

Setup: Specify which directories to expose (sandboxes access):

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

Best for: Any workflow where you want your AI to read existing code, config files, or docs without manual copy-paste.

2. github — Full GitHub Access

Source: Official Anthropic | Install: npx @modelcontextprotocol/server-github

This is the one that changes workflows. With the GitHub MCP server, you can ask Claude to “create a PR for this change,” “comment on issue #142,” “check what CI is failing,” or “list all open issues assigned to me” — without touching the GitHub UI.

What it can do:

• Create, read, update repositories
• File and comment on issues and PRs
• Search code across your org
• Read and create GitHub Actions workflows
• Manage branches, commits, releases

Setup: Requires a GitHub Personal Access Token with the repo scope.

Best for: Developers who want AI that can actually close the loop on GitHub tasks, not just suggest code.

3. postgres — Direct Database Queries

Source: Official Anthropic | Install: npx @modelcontextprotocol/server-postgres

Describe what you want in plain English, Claude writes and runs the SQL. The postgres server can read your schema, execute queries, and return results — turning natural language into data analysis.

Setup: Pass your database connection string:

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

Best for: Data analysis, debugging production issues, writing complex queries with schema context.

Note: For production databases, use a read-only connection string unless you explicitly want write access.

4. brave-search — Real-Time Web Search

Source: Official Anthropic | Install: npx @modelcontextprotocol/server-brave-search

Gives your AI access to current web search results — not training data from a year ago, but what’s actually on the web right now. Also includes news search.

Setup: Requires a free Brave Search API key.

Best for: Research tasks, checking if docs/APIs have changed, finding current pricing or product info.

5. puppeteer — Browser Automation

Source: Official Anthropic | Install: npx @modelcontextprotocol/server-puppeteer

Launches a real headless browser controlled by your AI. It can navigate to any URL, click buttons, fill forms, take screenshots, and extract content from JavaScript-heavy pages.

What it can do:

• Screenshot any webpage
• Click through multi-step web flows
• Extract data from SPAs (JavaScript-rendered content)
• Fill and submit forms
• Test your own web app

Best for: Web scraping, automated testing, research that requires navigating pages, verifying UI changes.

6. memory — Persistent Knowledge Graph

Source: Official Anthropic | Install: npx @modelcontextprotocol/server-memory

MCP sessions don’t persist by default — each new conversation starts blank. The memory server maintains a knowledge graph that persists between sessions. Your AI can store facts, relationships, and context that carry forward.

Use cases:

• Remember your project architecture decisions
• Track people, orgs, and relationships in research
• Store preferences and context across sessions
• Build up a knowledge base that grows with use

Best for: Power users who want Claude to actually remember things.

7. sequential-thinking — Better Reasoning

Source: Official Anthropic | Install: npx @modelcontextprotocol/server-sequentialthinking

Forces Claude to reason through problems step-by-step using a structured thought process, with the ability to revise earlier steps. For complex problems, this consistently outperforms default reasoning.

Best for: Architecture decisions, debugging complex issues, any task where reasoning quality matters more than speed.

8. fetch — HTTP Requests

Source: Official Anthropic | Install: npx @modelcontextprotocol/server-fetch

Makes HTTP requests from within your AI session. Read any URL, call any API, check API documentation, verify endpoints. Also intelligently converts HTML to markdown so Claude can read web pages cleanly.

Best for: Reading API docs in-context, testing your own APIs, fetching any web content.

9. supabase — Full Supabase Stack

Source: Community | Install: npm install @supabase/mcp-server-supabase

Goes beyond raw SQL — gives Claude access to Supabase’s full management API: database tables, storage buckets, edge functions, auth users, and project settings.

Best for: Supabase users who want AI that can read schema, write migrations, manage storage, and query data without leaving the AI interface.

10. linear — Project Management

Source: Community | Install: Via Linear’s official integration

Query and create Linear issues, update statuses, manage cycles, and search your project backlog from Claude. Great for dev workflows where you want AI to create tickets for bugs it finds.

Best for: Engineering teams using Linear who want their AI coding assistant to also manage the work queue.

Category Deep Dives

Databases

Developer Tools

Productivity

How to Set Up MCP in Claude Desktop

Step 1: Find your config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Step 2: Add servers to mcpServers:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/projects"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here" }
    },
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-brave-search"],
      "env": { "BRAVE_API_KEY": "your_brave_api_key" }
    }
  }
}

Step 3: Restart Claude Desktop. You’ll see the MCP tools available in new conversations.

Recommended Starter Stack

For developers:

• filesystem + github + brave-search + postgres (or sqlite)

For power users:

• filesystem + brave-search + memory + fetch + sequential-thinking

For Supabase projects:

• supabase + github + brave-search + sequential-thinking

For research:

• brave-search + fetch + memory + sequential-thinking

What’s Coming in MCP

The MCP ecosystem is growing fast. Expect to see:

• One-click installs — more servers packaged for frictionless setup
• Hosted MCP — server-side MCP with OAuth, no local install required
• More official servers — Anthropic is adding new reference implementations monthly
• Marketplace — centralized discovery is coming; directories like mcp-servers.directory are emerging

Have an MCP server that should be on this list? Suggest it here

Browse all AI developer tools → AI Coding Tools