What are the best MCP servers for developers?
MCP servers are most powerful for developers because developers already use tools with APIs: GitHub, databases, terminals, documentation systems, monitoring services. Each of these has an MCP server, and connecting them to your AI assistant changes how you work.
Here are the most useful ones, organized by what they do. For setup instructions, see How to install an MCP server.
Code and version control
GitHub MCP server The most widely used MCP server for developers. Manages repositories, issues, PRs, and file changes. Works best with Claude Code and Cursor. See the full GitHub MCP server guide for setup.
GitLab MCP server Same capabilities as GitHub but for GitLab-hosted repos. Useful if your team uses GitLab for enterprise or self-hosted reasons.
Git MCP server Works with local git repos without an API key. Read local commit history, show diffs, stage and commit changes. Essential for offline or self-hosted workflows.
Databases
Postgres MCP server Run natural language queries against any Postgres database. Especially useful for debugging production issues without writing SQL. See the database MCP server guide for setup.
SQLite MCP server For local development databases. Zero config once you point it at the file.
BigQuery MCP server For teams running analytics on GCP. The AI can write complex queries against your data warehouse.
Terminal and file system
Filesystem MCP server Read, write, and search files on your local machine. Combined with an AI coding assistant, this lets the AI actually work with your project files.
Shell MCP server Runs shell commands and returns output. Useful for build pipelines and scripting, with obvious security caveats (only run this in trusted environments).
Web and search
Brave Search MCP server Real-time web search. Gives your AI assistant current information beyond its training cutoff. Useful for looking up documentation, checking package versions, or researching APIs.
Fetch MCP server Fetches any URL and returns the content. The AI can read documentation, API responses, and web pages.
Puppeteer MCP server Runs a headless browser. Can screenshot pages, fill out forms, and scrape dynamic sites. Useful for UI testing workflows.
Monitoring and observability
Sentry MCP server Connects to Sentry error tracking. Ask the AI "what are the top 5 errors this week?" and get a summary with stack traces.
Datadog MCP server Queries Datadog metrics and logs. Useful for incident response: ask the AI to summarize what happened during a spike.
Productivity
Notion MCP server Connect your Notion workspace to search documents, create pages, and update databases. See the Notion MCP server guide for setup.
Slack MCP server Search message history, send updates to channels, and pull channel context. See the Slack MCP server guide for setup.
How do I choose which servers to start with?
Start with the tools you already use daily. If you live in GitHub and Postgres, start there. Add servers gradually as you identify friction points in your workflow.
Local setup: Most servers above run locally via npx (Node.js) or pip install (Python). A few (like Notion) also offer remote servers that connect via a URL with no installation needed. Check each listing on MCP Marketplace to see which modes are supported.
Check security scores on MCP Marketplace before installing any server. The score reflects how well the server handles your credentials and what permissions it requests.
Browse the full developer server list on MCP Marketplace.