Configure an MCP client manually
By hand
Section titled “By hand”Claude Code
claude mcp add kurlo -- npx -y @kurlo/cli mcp --env LocalCodex (CLI, IDE extension and desktop app share one config)
codex mcp add kurlo -- npx -y @kurlo/cli mcp --env Localor in ~/.codex/config.toml:
[mcp_servers.kurlo]command = "npx"args = ["-y", "@kurlo/cli", "mcp", "--env", "Local"]Claude Desktop, Cursor and other clients with a JSON config (claude_desktop_config.json, .cursor/mcp.json, …)
{ "mcpServers": { "kurlo": { "command": "npx", "args": ["-y", "@kurlo/cli", "mcp", "--env", "Local"] } }}Without Node.js
Section titled “Without Node.js”The desktop app is the same server. Point the client at its binary instead of npx: /Applications/Kurlo.app/Contents/MacOS/kurlo on macOS, or the installed kurlo executable on Linux and Windows.
claude mcp add kurlo -- /Applications/Kurlo.app/Contents/MacOS/kurlo mcp --env LocalWith no workspace argument, the server uses the workspace the app has open, including the secret values the app keeps on this machine. To serve a different Git-backed workspace, pass its folder (the one that contains kurlo.yml) as the first argument:
kurlo mcp ~/code/payments-api/kurlo --env StagingThe workspace is read again on every call, so a request you edit in the app is what the assistant runs next. In the other direction, the app notices files the assistant wrote and reloads them. If you were editing at that moment, both changes are kept: see Working alongside the app.