Filesystem
Files & DocsRead and write files in directories you allow-list.
GitHub
CodeRead repos, manage issues, open PRs from inside Claude or Cursor.
Puppeteer
BrowserNavigate, click, screenshot, and scrape live web pages.
Postgres
DataQuery Postgres databases with SQL the model writes for you.
SQLite
DataQuery and modify SQLite databases on disk.
Brave Search
SearchWeb search results piped into the model, free tier 2k/month.
Slack
APIRead channels, post messages, manage threads from Claude.
Fetch
BrowserFetch a URL and return clean text, the lightweight Puppeteer.
Memory
ProductivityPersistent knowledge graph the model can recall across sessions.
Sentry
DevOpsPull error events and issue context into Claude for triage.
Categories explained
FAQ
What is MCP?
Model Context Protocol is Anthropic's open standard for letting AI clients (Claude Desktop, Cursor, etc.) talk to external systems. Each server exposes tools the model can call.
How do I install one of these in Claude Desktop?
Edit ~/Library/Application Support/Claude/claude_desktop_config.json (Mac) or the Windows equivalent, paste the config JSON shown on the server's detail page, then quit and relaunch Claude Desktop. The config is read once at startup.
Will Cursor configs be different?
Mostly the same shape. Cursor uses ~/.cursor/mcp.json and supports the same protocol. Detail pages note when the config differs.
What if a server isn't here?
We start with 10 servers we can describe in depth and verify in real workflows. We expand the list every two weeks based on traction. The official Anthropic monorepo at github.com/modelcontextprotocol/servers has the full upstream list.
Are these safe to install?
All Anthropic-published servers are sandboxed by client permission scopes. Community servers vary — read the source repo before installing one with elevated permissions (writes, secrets, etc.).