Must-Have MCP Servers for Every Developer

You just asked your AI to refactor a React component, but it generated code using a hook that was deprecated two versions ago. You asked it to write a migration, but it guessed at your table schema instead of reading it. You asked it to update a Jira ticket, but you had to copy-paste the ticket contents into chat first. Every one of these friction points has the same root cause: your AI is missing context that exists in a tool you already use.

The right MCP servers eliminate these problems permanently. Not by adding complexity, but by giving your AI direct access to the information it needs to do its job.

What You’ll Walk Away With

• A curated list of five essential MCP server categories with specific installation instructions for all three tools
• Copy-paste configuration blocks for Cursor, Claude Code, and Codex
• A prioritized setup order that delivers maximum value with minimum effort
• Prompts that demonstrate the real capability of each server category

The Foundation: Five Server Categories

1. Version Control — GitHub MCP Server

The single most impactful MCP server for any team using GitHub. It gives your AI the ability to search code across repositories, read pull request diffs, check CI status, and manage issues — all without you pasting links into chat.

Cursor

.cursor/mcp.json

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

Claude Code

.claude/settings.json

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

Codex

~/.codex/config.toml

[mcp.github]
transport = "stdio"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-github"]

[mcp.github.env]
GITHUB_PERSONAL_ACCESS_TOKEN = "<your-token>"

Tip

Copy-paste prompt for GitHub MCP:

“Search the main branch of our repository for all files that import the useAuth hook. Then check the last 5 pull requests that modified those files and summarize what changed in the authentication flow.”

2. Filesystem — Official Filesystem MCP

The filesystem server gives your AI structured access to read and write files within allowed directories. Without it, your AI is limited to the files currently open in your editor. With it, the AI can explore your entire project, create new files, and apply changes across multiple files at once.

Cursor

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

Claude Code

Claude Code already has built-in filesystem access through its native tools, so a separate filesystem MCP server is typically unnecessary. Claude Code can read, write, and search files directly.

Codex

[mcp.filesystem]
transport = "stdio"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/your/project"]

Caution

Always scope the filesystem server to specific directories. Never point it at your home directory or root. The allowed_directories argument ensures the AI cannot access files outside your project.

3. Up-to-Date Documentation — Context7

Language models have a knowledge cutoff. Context7 solves this by fetching current documentation for thousands of open-source libraries on demand. When your AI needs to use an API, it pulls the latest docs instead of relying on training data that might be months old.

Cursor

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

Claude Code

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

Codex

[mcp.context7]
transport = "stdio"
command = "npx"
args = ["-y", "@upstash/context7-mcp@latest"]

Tip

Copy-paste prompt for Context7:

“Use context7 to look up the Drizzle ORM documentation for defining relations between tables. Then create a one-to-many relationship between my users and posts tables using the patterns from the current docs.”

4. Browser Automation — Playwright MCP

Playwright MCP gives your AI control over a headless browser. It can navigate pages, click buttons, fill forms, take screenshots, and read the accessibility tree — all without expensive screenshot-based vision. This transforms testing workflows: the AI can verify its own changes by actually running the app.

Cursor

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

Claude Code

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

Codex

[mcp.playwright]
transport = "stdio"
command = "npx"
args = ["-y", "@playwright/mcp@latest"]

Tip

Copy-paste prompt for Playwright MCP:

“Navigate to localhost:3000/login, fill in the email field with ‘test@example.com’ and the password field with ‘testpass123’, click the submit button, and verify that the page redirects to /dashboard. Take a screenshot of the final page state.”

5. Project Management — Atlassian or Linear

Connecting your project management tool closes the loop between coding and coordination. The AI can read ticket requirements before writing code, update ticket status when work is complete, and even create new issues for bugs it discovers.

For Atlassian (Jira + Confluence), use the official Rovo MCP server — it is a remote server that connects via OAuth:

Cursor

{
  "mcpServers": {
    "atlassian": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.atlassian.com/v1/mcp"]
    }
  }
}

Claude Code

{
  "mcpServers": {
    "atlassian": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.atlassian.com/v1/mcp"]
    }
  }
}

Codex

[mcp.atlassian]
transport = "http"
url = "https://mcp.atlassian.com/v1/mcp"

Tip

Copy-paste prompt for project management MCP:

“Find all open bugs assigned to me in the PLATFORM project. For the highest-priority one, read the description and acceptance criteria, then create a feature branch named after the ticket ID and start implementing the fix.”

The Right Installation Order

1. Context7 first. Zero configuration, immediate value. Every prompt you write benefits from up-to-date documentation.

2. GitHub MCP second. Requires a personal access token but transforms how the AI understands your codebase history and team activity.

3. Playwright third. No API keys needed. Gives the AI the ability to verify its own work in the browser, catching visual regressions before you do.

4. Filesystem if needed. Cursor and Codex benefit from this; Claude Code already has native file access.

5. Project management last. Requires OAuth setup but completes the development lifecycle — from ticket to code to deployment.

When This Breaks

GitHub MCP returns “Bad credentials”. Your personal access token needs repo, read:org, and read:user scopes at minimum. Generate a new fine-grained token with the correct permissions.

Context7 returns empty results. Not every library is indexed. If Context7 does not have docs for your library, fall back to the Fetch MCP server to grab documentation pages directly.

Playwright cannot connect to localhost. Make sure your dev server is running before invoking the browser tools. The Playwright MCP server does not start your app — it just connects to whatever is already listening on the specified port.

Atlassian OAuth flow fails. The first user on your Atlassian site must have access to both Jira and Confluence for the initial OAuth consent. Subsequent users can authorize with access to just one product.

What’s Next

Filesystem MCP Deep Dive Advanced file operations, directory scoping, and security patterns.

Browser Automation MCP Playwright and Puppeteer MCP for testing and web interaction.

Database MCP Servers Connect your AI to PostgreSQL, MongoDB, SQLite, and more.

Version Control MCP GitHub and GitLab MCP integration deep dive.