Thousands of MCP servers exist. These 15 are the most useful for developers — tested, maintained, and ready to use with Claude Code, Cursor, and other MCP hosts.

Development tools

1. GitHub — Read repos, create issues, manage PRs, search code.

claude mcp add github npx @modelcontextprotocol/server-github

2. Filesystem — Read, write, and search local files.

claude mcp add fs npx @modelcontextprotocol/server-filesystem /path/to/dir

3. Git — Log, diff, blame, branch management.

Databases

4. PostgreSQL — Query your database from AI conversations.

5. SQLite — Local database for development.

6. Redis — Cache operations, key management.

Communication

7. Slack — Send messages, read channels, search history.

8. Email — Send/read via IMAP/SMTP.

Web & search

9. Brave Search — Web search for building AI search.

10. Fetch — Extract content from any URL.

Cloud & infrastructure

11. AWS — Manage AWS resources. See our AWS CLI cheat sheet.

12. Docker — Container management.

13. Kubernetes — Cluster management.

AI & data

14. Ollama — Connect to local models.

15. Vector DB — Query vector databases for RAG.

Installation

# Claude Code
claude mcp add <name> npx <package>

# Cursor: Settings → MCP → Add Server

Build your own: TypeScript tutorial · Python tutorial

Review security before installing: MCP Security Checklist

How to choose MCP servers

Not every MCP server is worth installing. Here’s how to evaluate them:

Maintenance status — Check the GitHub repo. Is it actively maintained? MCP is evolving fast, and abandoned servers break with protocol updates.

Permission scope — Does the server request more access than it needs? A GitHub MCP server shouldn’t need filesystem access. Review permissions before installing.

Error handling — Good MCP servers fail gracefully. Bad ones crash your AI session when an API call times out. Test with edge cases before relying on them.

Rate limiting — Servers that call external APIs (GitHub, Slack, databases) can hit rate limits during long AI sessions. Check if the server handles this or if it just throws errors.

Security considerations

MCP servers run locally with your permissions. A malicious server could:

• Read your filesystem (including SSH keys and credentials)
• Make network requests on your behalf
• Modify or delete files

Only install servers from trusted sources. Review the code before running anything that requests broad filesystem or network access. See our MCP Security Checklist for a full audit guide.

Building custom MCP servers

The most useful MCP servers are often the ones you build yourself — tailored to your specific workflow:

// Example: Custom MCP server for your project's API
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";

const server = new McpServer({ name: "my-api", version: "1.0.0" });

server.tool("get-user", { id: z.string() }, async ({ id }) => {
  const user = await db.users.findById(id);
  return { content: [{ type: "text", text: JSON.stringify(user) }] };
});

This lets your AI assistant query your production database, check deployment status, or interact with internal tools — all through natural language.

See our full tutorials: TypeScript MCP server · Python MCP server

MCP server performance tips

1. Don’t install too many — Each server adds startup time and memory. 5-8 focused servers beats 20 rarely-used ones.
2. Use stdio transport for local servers — It’s faster and more reliable than HTTP for same-machine communication.
3. Cache expensive operations — If your server queries an API, cache results to avoid redundant calls during long sessions.

FAQ

What are the best MCP servers for developers?

The most essential MCP servers are GitHub (for repo management and PR workflows), Filesystem (for file operations), and your database server (PostgreSQL or SQLite). These three cover the majority of developer workflows. Add Brave Search for web lookups and Docker for container management as needed.

Are MCP servers safe to install?

MCP servers run with your local permissions, so you should treat them like any other code you run on your machine. Only install from trusted sources, review the permission scope, and check the GitHub repo for active maintenance. Our MCP Security Checklist covers the full audit process.

How many MCP servers should I use?

Keep it to 5-8 focused servers. Each server adds startup time and memory overhead. Install only what you actively use — it’s easy to add more later. Too many servers can also confuse the AI about which tool to use for a given task.

Can I build my own MCP server?

Yes, and it’s straightforward. The MCP SDK is available for TypeScript and Python. Custom servers are often the most useful because they’re tailored to your specific workflow — querying your database, checking your CI pipeline, or interacting with internal APIs.

Related: What is MCP? · MCP Complete Guide · MCP + Claude Code Setup