wendy cloud mcp
Let a web chat (ChatGPT, Claude) use Wendy through this session
Syntax
wendy cloud mcp [flags]Examples
wendy cloud mcp --helpArguments
No positional arguments are declared in the command synopsis.
Flags
Flags below include inherited options. Types and defaults come from the CLI source. An empty default means no value is configured.
| Flag | Type | Default | Description |
|---|---|---|---|
--allow-all | bool | false | Allow every tool without per-call confirmation; does not change pairing approval |
--allow-read | bool | false | Automatically approve read-only device and Cloud tools that do not access the host |
--allow-tool | stringSlice | [] | Also allow these tools by name (comma-separated); state changes and host access require confirmation |
--auto-accept | bool | false | Automatically accept ChatGPT pairing; tool permissions and per-call confirmations are unchanged |
--cloud-grpc | string | (empty) | Cloud gRPC endpoint of the session to use |
--device | string | (empty) | Target device host (hostname, IP, or host:port); 'wendy run' accepts a comma-separated list to deploy one build to several devices (needs a remote build host and --detach). Defaults to $WENDY_DEVICE, then the saved default device |
--json | bool | false | Output in JSON format |
Output
Output depends on the operation and selected target. Command behavior and detailed examples.
JSON output
The global --json option requests machine-readable output. The CLI also selects JSON mode when no interactive terminal is attached. Support and payloads are command-specific; this flag does not guarantee a single JSON document for commands that stream logs or invoke build tools. Consult Details before parsing command output.
Exit codes
The CLI entry point returns 0 on success and for its user-cancelled or default-cleared results. Other returned errors exit with 1 and print a diagnostic to stderr. Some operations may terminate through subprocess or signal handling; do not infer application health from the CLI exit code alone.
Common errors
Check argument values and flag combinations against the synopsis above. For discovery, build, port, and readiness failures, see deployment troubleshooting. Command-specific constraints are in Details.
See also
Details
Open a session with the hosted Wendy MCP relay and serve the wendy MCP tools to a web chat you pair with it. Every call runs here and is signed with your operator key; the chat can do nothing once this command exits. Every permitted tool call requires confirmation by default. Use --allow-read to automatically approve read-only device and Cloud tools that do not access the host. Use --allow-tool to permit other tools with confirmation, or --allow-all to permit every tool without per-call confirmation. Pairing requires your acceptance unless --auto-accept is set for ChatGPT. That flag only accepts ChatGPT's verified OAuth client ID; other clients still need confirmation. A terminal is required for interactive confirmations.