Best MCP Servers for Cursor: The Complete 2026 Setup Guide

Cursor is an AI-first code editor with native MCP support, which means it can connect to any MCP server the same way Claude Desktop or Claude Code does. The difference is what you’d actually want connected while you’re heads-down in an editor: file operations, git, a database, maybe a browser for testing.

This guide covers how Cursor connects to MCP servers, which ones are worth installing, and how to set them up.

How Cursor Connects to MCP Servers

Cursor reads MCP server configuration from two places:

• Project-level: .cursor/mcp.json in your repo root — scoped to that project, shareable with your team via version control
• Global: Cursor’s settings panel under MCP — available across every project you open

Most servers connect over stdio, since Cursor runs locally and the server runs as a child process on the same machine. This is the simplest setup: no network config, no auth server to stand up. Remote or team-shared servers use SSE and need a URL plus whatever auth the server requires.

A minimal config looks like this:

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

Top MCP Servers for Cursor

Server	What it does	Transport	Auth	Best for
Filesystem	Read/write/edit files beyond Cursor’s built-in reach	stdio	None	Multi-repo or cross-directory work
GitHub	Issues, PRs, commits, repo search	stdio/SSE	Personal access token	Teams that live in GitHub workflows
Postgres / SQLite	Query and inspect databases directly	stdio	Connection string	Backend and data work without leaving the editor
Playwright	Browser automation and testing	stdio	None	End-to-end test generation and debugging
Brave Search	Live web search	stdio	API key	Looking up current docs or library versions mid-session
Context7	Up-to-date library and framework documentation	stdio	None	Avoiding outdated training-data API calls
Linear	Issue tracking	SSE	OAuth2	Teams that plan and code in the same loop
Sentry	Error monitoring and stack traces	SSE	API key	Debugging production issues without tab-switching

Setting One Up: GitHub as an Example

1. Get a token. Generate a GitHub personal access token with repo scope.

2. Add it to your config. Drop this into .cursor/mcp.json:

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

3. Restart Cursor (or reload the MCP panel) so it picks up the new server.

4. Verify the connection. Cursor’s MCP settings panel shows a green status indicator once the server responds.

5. Never commit the token. Use environment variables or a .env file excluded from git — project-level configs get shared, and a hardcoded token in .cursor/mcp.json ends up in your repo history.

FAQ

Does Cursor support all MCP transports?

Yes. stdio for local servers, SSE for remote ones. Cursor doesn’t restrict which transport a server uses.

Can I use the same MCP servers in Cursor and Claude Desktop?

Yes. MCP servers aren’t client-specific. The same server config works across any MCP-compatible host, though each client has its own config file location and format.

Do project-level and global configs conflict if I have the same server in both?

Project-level takes precedence for that project. Use global config for servers you want everywhere (filesystem, search) and project-level for anything repo-specific (a database connection string tied to one codebase).

How many MCP servers should I connect at once in Cursor?

Keep it lean. Each connected server’s tools load into context before you type anything. Four to six purpose-built servers beats one server bundling everything — same principle as any other MCP client.

---

Browse the full catalog at MyMCP Shelf.