Where to Find Cursor IDE MCP Settings (mcp.json Path) — 2026 Guide

Where are Cursor IDE MCP settings? In the mcp.json file inside your home .cursor folder — ~/.cursor/mcp.json on Mac and Linux, %USERPROFILE%\.cursor\mcp.json on Windows. Open it, add an MCP server under mcpServers, save, and restart Cursor (or reload MCP) so the new server appears.

This guide owns the Cursor IDE MCP setup / where to find mcp.json query for 2026: config paths, how to add servers, and the errors that show up when the path or JSON is wrong. Product-specific installs (GitHub, Jira, Confluence, Linear, and others) have their own dedicated guides linked below.

If you want the same connected tools reachable outside the editor entirely — from a global keyboard shortcut rather than an open Cursor window — the Raycast MCP Server Setup guide covers the launcher-side equivalent, config file and all.

Where Is the Cursor MCP Config File?

This is the most common question, so let's answer it first.

Cursor stores MCP configuration in a file called mcp.json inside the .cursor directory in your home folder. The exact path depends on your operating system:

To open it directly:

# macOS / Linux
open ~/.cursor/mcp.json
# or edit with VS Code
code ~/.cursor/mcp.json

# Windows (PowerShell)
notepad "$env:USERPROFILE\.cursor\mcp.json"

To open it from inside Cursor:

1. Press Cmd/Ctrl + Shift + P to open the command palette
2. Type Open MCP Settings and select it
3. Cursor opens mcp.json directly in the editor

Note: If the .cursor folder or mcp.json doesn't exist yet, create it. Cursor will pick it up on the next restart.

> Project-scoped vs global config: You can also place an mcp.json inside a specific project's .cursor/ folder. Project-level config takes precedence over the global config for that workspace. Global config (in your home directory) applies everywhere.

Prerequisites

Before adding any MCP servers:

The mcp.json File Format

The config file uses a straightforward JSON structure:

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

Each key inside mcpServers is a label you choose — it's what Cursor calls the server internally. You can add as many servers as you need.

Quick Start: Add GitHub MCP in 3 Minutes

GitHub is the most useful first MCP server for most developers. It lets Cursor read issues, PRs, and repo structure.

Step 1: Get a GitHub Personal Access Token

Step 2: Open your ~/.cursor/mcp.json and add:

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

Step 3: Restart Cursor completely (quit and reopen — not just reload window)

Step 4: Open Cursor chat (Cmd/Ctrl + L) and try:

Popular MCP Servers and Their Configs

PostgreSQL / Database

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": {
        "POSTGRES_CONNECTION_STRING": "postgresql://user:pass@localhost/dbname"
      }
    }
  }
}

Try asking: "Show me all users who signed up in the last 30 days"

Filesystem (Full Project Access)

Gives Cursor read access to any directory — useful for large codebases or mono-repos.

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

Try asking: "Find all files that import the AuthService"

Context7 (Up-to-date Library Docs)

Context7 fetches current documentation for any library directly into Cursor's context — no more hallucinated APIs.

{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp"]
    }
  }
}

Try asking: "Use context7 to get the latest Next.js 15 App Router docs"

Atlassian (Jira + Confluence)

{
  "mcpServers": {
    "atlassian": {
      "command": "npx",
      "args": ["-y", "@atlassian/mcp-atlassian"],
      "env": {
        "ATLASSIAN_API_TOKEN": "your_api_token",
        "ATLASSIAN_SITE_URL": "https://yourcompany.atlassian.net",
        "ATLASSIAN_USER_EMAIL": "you@company.com"
      }
    }
  }
}

Try asking: "What Jira tickets are assigned to me?" or "Find the API documentation page in Confluence"

This one config block covers both products, but each has its own deeper guide: see the Jira MCP server setup for ticket/sprint workflows and the Confluence MCP server setup for space search, page creation, and attachments.

Brave Web Search

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

Try asking: "Search for the current rate limits for OpenAI's API"

For the full setup including free-tier limits, local search, and troubleshooting, see the Brave Search MCP server guide.

Running Multiple MCP Servers

You can run as many servers as you want simultaneously. Just add each one as a key in mcpServers:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..."
      }
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": {
        "POSTGRES_CONNECTION_STRING": "postgresql://..."
      }
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/your/project"]
    }
  }
}

Cursor starts each server in a separate process. Servers that fail to start don't affect the others.

Using the cwd Field

Some servers need to know which directory to treat as the working directory:

{
  "mcpServers": {
    "my-server": {
      "command": "python",
      "args": ["server.py"],
      "cwd": "/Users/you/projects/my-mcp-server"
    }
  }
}

This is important when the server script uses relative file paths internally.

Troubleshooting Common Errors

"MCP server not found" or server doesn't appear in Cursor

1. Check your JSON is valid — use jsonlint.com to paste and validate
2. Use absolute paths in args and cwd, not ~/ shortcuts (Cursor may not expand them)
3. Fully restart Cursor — closing the window isn't always enough; quit from the menu bar
4. Check that npx is in your PATH — run which npx in terminal; if blank, reinstall Node.js

"Tool not available" in Cursor chat

Server starts but gives wrong data / stale responses

Environment variables not loading

If your server needs API keys, put them directly in the env block rather than relying on your shell's .env or .bashrc — Cursor doesn't always inherit shell environment variables.

{
  "env": {
    "MY_API_KEY": "actual-key-value-here"
  }
}

Build a Custom MCP Server

If you need a server that doesn't exist yet, building one is straightforward.

from mcp.server import Server
from mcp.types import Tool

app = Server("my-custom-server")

@app.tool()
async def search_internal_docs(query: str) -> str:
    """Search your company's internal documentation.
    
    Args:
        query: What to search for
    
    Returns:
        Relevant documentation excerpts
    """
    # Your search logic here
    results = my_search_function(query)
    return "\n".join(results)

if __name__ == "__main__":
    import asyncio
    asyncio.run(app.run_stdio())

Install the SDK: pip install mcp

The docstring is what Cursor's AI reads to understand when and how to use the tool — write it like you're explaining to a smart colleague who doesn't know your codebase.

Quick Reference

Related Guides