How it works
- Enter a server name and pick the transport. Streamable HTTP needs a URL and, if the server asks for one, a bearer token. Stdio needs the command that starts the server.
- The page builds each config in your browser as you type. Nothing you type is sent to our server, and the token is never put in the page URL.
- Copy the block for your client into the file named above it, or run the command in a terminal. Then restart the client or reload its MCP servers.
- Where a client cannot take a remote server with headers in its config file, the page says so instead of making a config that does not work.
FAQ
Frequently asked questions
Is my token sent anywhere?
No. The configs are built by JavaScript in your browser. The token is not sent to our server, not stored and not added to the shareable link. It is hidden by default.
Which transport should I pick?
Pick Streamable HTTP when the server already runs and gives you a URL, like the Focus Pad server at http://127.0.0.1:47821/mcp. Pick stdio when the client must start the server itself with a command such as npx or uvx.
Why is there no Claude Desktop file for an HTTP server?
The Claude Desktop config file is documented for local stdio servers. You add a remote server with a header in the app: Customize, Connectors, Add custom connector.
Why does Codex use an environment variable for the token?
The Codex docs read a bearer token from the environment variable named in bearer_token_env_var. The page gives you the export line for your shell and keeps the token out of config.toml.
Where does each file go?
The path is shown above each block. Claude Code stores user-scope servers itself when you run the command. Cursor uses ~/.cursor/mcp.json, Codex ~/.codex/config.toml, OpenCode opencode.json and VS Code .vscode/mcp.json.
DevWorkspace Pro
One workspace for every project
Files, Git, GitHub, AI agents, the terminal and your runtimes in one desktop app. Works with every project folder.