Set up MCP servers in your AI client
Step-by-step guides for adding Model Context Protocol servers to the clients people actually use, written from each client's official documentation and illustrated with real examples from this registry.
Guides
- Add MCP servers to Claude Code and Codex CLI
Register servers from the terminal with claude mcp add and codex mcp add, choose a scope, sign in to OAuth servers and check that they connected.
- Add MCP servers to Cursor
Configure ~/.cursor/mcp.json or a project .cursor/mcp.json, connect OAuth servers, keep secrets in environment variables and control tool approval.
- Add MCP servers to VS Code
Use .vscode/mcp.json or your user configuration, prompt for secrets with inputs, understand server trust and sandboxing, and debug with the MCP output log.
- Add MCP servers to Claude Desktop
Connect remote servers as custom connectors, add local servers in claude_desktop_config.json, and find the MCP logs when something fails.
- Remote vs local MCP servers
How stdio and Streamable HTTP servers differ in setup, access, authentication and risk, with the registry's current servers sorted by transport.
- MCP authentication and security basics
OAuth versus API keys, least-privilege tokens, keeping secrets out of shared files, reviewing tool calls, and the risks the MCP specification warns about.
What the registry covers today
The registry currently lists 33 current, configurable servers. 21 are remote servers you connect to by URL: 12 of them sign you in through OAuth, 7 need no authentication at all and 2 take a token or key in a header. The other 12 run locally as a process your client starts. Every listing shows a configuration example for Claude Code, Cursor, Codex CLI, Gemini CLI, OpenCode and VS Code; the guides explain where each one goes and what to check first.
Not sure where to start? Read remote vs local servers, then the guide for your client, then authentication and security basics before you connect anything to an account that matters.