The best MCP servers for developers in 2026 are Filesystem (local file access), GitHub (repo management via Docker), Brave Search (web search), PostgreSQL (database queries), Memory (persistent context), Puppeteer (browser automation), Slack (team messaging), and SQLite (lightweight databases). Most run with a single npx command and work across Claude Desktop, Cursor, VS Code, and Windsurf. This curated list covers the most useful servers with installation commands and configuration examples.
Curated List of the Most Useful MCP Servers for Developers
The MCP ecosystem has grown to over 10,000 servers, making it hard to know where to start. This guide curates the most useful, reliable, and well-maintained servers for common developer workflows. Each server is listed with its installation command, tool count, and recommended configuration. These are battle-tested servers used by thousands of developers daily.
Prerequisites
- An MCP-compatible host installed (Claude Desktop, Cursor, VS Code, or Windsurf)
- Node.js 18+ installed for npx-based servers
- Docker installed for container-based servers (optional, only for GitHub server)
Step-by-step guide
Set up the Filesystem server for local file access
Set up the Filesystem server for local file access
The Filesystem server by Anthropic gives AI assistants read and write access to files and directories on your local machine. It exposes tools for reading files, writing files, listing directories, searching file contents, and creating directories. You specify which directories the server is allowed to access, providing a security boundary. This is the most commonly used MCP server and the first one most developers install.
1// Configuration for any host using mcpServers key:2"filesystem": {3 "command": "npx",4 "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/projects"]5}67// Test with Inspector:8// npx -y @modelcontextprotocol/inspector npx -y @modelcontextprotocol/server-filesystem /Users/you/projects910// Tools: read_file, write_file, list_directory, search_files, create_directory, move_file, get_file_info11// ~11 tools totalExpected result: The AI can read, write, and search files in the specified directory.
Set up the GitHub server for repository management
Set up the GitHub server for repository management
The GitHub MCP server provides tools for managing repositories, issues, pull requests, branches, and file contents via the GitHub API. It runs as a Docker container (the npm package was deprecated in April 2025). You need a GitHub Personal Access Token with appropriate permissions. This server exposes 20+ tools covering most GitHub operations.
1// Configuration (requires Docker):2"github": {3 "command": "docker",4 "args": ["run", "-i", "--rm",5 "-e", "GITHUB_PERSONAL_ACCESS_TOKEN",6 "ghcr.io/github/github-mcp-server"],7 "env": {8 "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here"9 }10}1112// Tools: create_issue, list_issues, create_pull_request, search_repositories,13// get_file_contents, push_files, create_branch, list_commits, and more14// ~20+ tools totalExpected result: The AI can manage GitHub repositories, issues, and pull requests.
Set up the Brave Search server for web search
Set up the Brave Search server for web search
The Brave Search server gives AI assistants the ability to search the web in real time. It uses the Brave Search API which provides high-quality results without tracking. You need a free Brave Search API key (2,000 queries/month on the free tier). This is essential for any workflow that needs current information beyond the model's training data.
1// Configuration:2"brave-search": {3 "command": "npx",4 "args": ["-y", "@modelcontextprotocol/server-brave-search"],5 "env": {6 "BRAVE_API_KEY": "your-brave-api-key"7 }8}910// Get a free API key at: https://brave.com/search/api/11// Tools: brave_web_search, brave_local_search12// ~2 tools totalExpected result: The AI can search the web for current documentation, solutions, and information.
Set up the PostgreSQL server for database queries
Set up the PostgreSQL server for database queries
The PostgreSQL server lets AI assistants query your database, inspect schemas, and generate data reports. It connects to any PostgreSQL database via a connection string. The server provides read-only access by default for safety. This is powerful for data exploration and query generation.
1// Configuration:2"postgres": {3 "command": "npx",4 "args": ["-y", "@modelcontextprotocol/server-postgres"],5 "env": {6 "POSTGRES_CONNECTION_STRING": "postgresql://user:password@localhost:5432/mydb"7 }8}910// Tools: query, list_tables, describe_table, list_schemas11// ~5 tools totalExpected result: The AI can query your PostgreSQL database and inspect table schemas.
Set up the Memory server for persistent context
Set up the Memory server for persistent context
The Memory server provides persistent key-value storage that survives across conversations. AI assistants can save information (user preferences, project context, decisions) and retrieve it later. This is useful for maintaining context across multiple chat sessions — the AI can remember what you discussed previously.
1// Configuration:2"memory": {3 "command": "npx",4 "args": ["-y", "@modelcontextprotocol/server-memory"]5}67// No API keys or environment variables needed8// Tools: create_entities, create_relations, search_nodes, open_nodes, delete_entities9// ~5 tools totalExpected result: The AI can store and retrieve persistent context across conversations.
Explore additional recommended servers
Explore additional recommended servers
Beyond the core servers above, several other servers are valuable for specific workflows. Puppeteer gives AI assistants browser automation capabilities for web scraping and testing. SQLite provides lightweight database access. Slack enables team messaging integration. Sentry connects error tracking data. For organizations needing custom MCP servers for internal tools, RapidDev specializes in building tailored MCP integrations.
1// Puppeteer (browser automation):2"puppeteer": {3 "command": "npx",4 "args": ["-y", "@modelcontextprotocol/server-puppeteer"]5}67// SQLite (lightweight database):8"sqlite": {9 "command": "npx",10 "args": ["-y", "@modelcontextprotocol/server-sqlite", "/path/to/database.db"]11}1213// Slack (team messaging — requires bot token):14"slack": {15 "command": "npx",16 "args": ["-y", "@modelcontextprotocol/server-slack"],17 "env": { "SLACK_BOT_TOKEN": "xoxb-your-token" }18}Expected result: You know about additional servers available for specific use cases.
Complete working example
1{2 "mcpServers": {3 "filesystem": {4 "command": "npx",5 "args": [6 "-y",7 "@modelcontextprotocol/server-filesystem",8 "/Users/you/projects"9 ]10 },11 "github": {12 "command": "docker",13 "args": [14 "run", "-i", "--rm",15 "-e", "GITHUB_PERSONAL_ACCESS_TOKEN",16 "ghcr.io/github/github-mcp-server"17 ],18 "env": {19 "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token"20 }21 },22 "brave-search": {23 "command": "npx",24 "args": ["-y", "@modelcontextprotocol/server-brave-search"],25 "env": {26 "BRAVE_API_KEY": "your-brave-api-key"27 }28 },29 "postgres": {30 "command": "npx",31 "args": ["-y", "@modelcontextprotocol/server-postgres"],32 "env": {33 "POSTGRES_CONNECTION_STRING": "postgresql://user:pass@localhost:5432/mydb"34 }35 },36 "memory": {37 "command": "npx",38 "args": ["-y", "@modelcontextprotocol/server-memory"]39 },40 "puppeteer": {41 "command": "npx",42 "args": ["-y", "@modelcontextprotocol/server-puppeteer"]43 }44 }45}Common mistakes
Why it's a problem: Installing the deprecated GitHub npm package
How to avoid: The npm-based GitHub MCP server was deprecated in April 2025. Use the Docker image: ghcr.io/github/github-mcp-server. You need Docker installed.
Why it's a problem: Giving the filesystem server access to your entire home directory
How to avoid: Limit file access to specific project directories. Giving access to ~ exposes sensitive files like .ssh, .env, and credentials. Use the most restrictive path that covers your needs.
Why it's a problem: Using production database credentials for the PostgreSQL server
How to avoid: Create a read-only database user for MCP access. Never use admin or write credentials unless you specifically need write access. The AI model could inadvertently modify data.
Why it's a problem: Forgetting API keys for servers that need them
How to avoid: Brave Search needs BRAVE_API_KEY, GitHub needs GITHUB_PERSONAL_ACCESS_TOKEN, Slack needs SLACK_BOT_TOKEN. Check each server's documentation for required environment variables.
Best practices
- Start with Filesystem and one other server — add more only as needed
- Use the most restrictive file paths for the Filesystem server
- Create read-only database credentials for PostgreSQL MCP access
- Get a free Brave Search API key for web search capability (2,000 queries/month free)
- Use Docker for the GitHub server — the npm package is deprecated
- Count total tools across servers if using Windsurf (100-tool limit)
- Test each server individually with MCP Inspector before combining them
- Keep API keys in the env field of your config, never in the args array
Still stuck?
Copy one of these prompts to get a personalized, step-by-step explanation.
What are the best MCP servers for a full-stack developer? I want file access, GitHub integration, database queries, and web search. Show me the configuration for each server in Claude Desktop format, including any required API keys and environment variables.
Recommend the top 5 MCP servers I should set up for my development workflow. I work with TypeScript, PostgreSQL, and GitHub. Show the complete configuration file I need and explain what each server adds to my AI assistant's capabilities.
Frequently asked questions
Are all these servers free?
The MCP servers themselves are free and open source. Some require API keys for the underlying services: Brave Search has a free tier (2,000 queries/month), GitHub needs a free PAT, and Slack needs a bot token from your workspace.
Which server should I install first?
Start with the Filesystem server. It requires no API keys, works immediately, and is the most universally useful — giving your AI assistant access to read and explore your project files.
Do I need Docker for any of these servers?
Only the GitHub MCP server requires Docker (since the npm package was deprecated). All other recommended servers run via npx without Docker.
How many of these servers can I run at once?
All of them, in most hosts. Claude Desktop and Cursor have no hard tool limit. Windsurf has a 100-tool limit. Running all 6 servers listed here would expose approximately 50 tools total, well within limits.
Where can I find more MCP servers?
The official MCP servers repository on GitHub lists all Anthropic-maintained servers. Community servers are published on npm (search for 'mcp-server'), and directories like mcp.run and glama.ai/mcp/servers catalog thousands of third-party servers.
Can RapidDev build custom MCP servers for our specific tools?
Yes. RapidDev builds custom MCP servers that integrate with internal APIs, proprietary databases, and company-specific tools. This is ideal when off-the-shelf servers do not cover your needs.
Talk to an Expert
Our team has built 600+ apps. Get personalized help with your project.
Book a free consultation