node-connect
Node Connect
Goal: fix one exact client against one exact Gateway, then prove that client's fresh connection.
1. Lock the target
Record the target environment/profile, OpenClaw binary, config/state root, Gateway URL/port, and service before changing anything.
- Use the named deployment wrapper/profile for every config, log, service, device, and node command.
- Never drift to a bare
openclaw, proof environment, or similarly named deployment. - If the global executable is stale, invoke the target through its owner.
- Verify status, config, logs, and process identity describe the same Gateway.
Do not mutate pairing or auth until the target is unambiguous.
2. Classify the client
Classify from the request and Gateway log before choosing commands:
- Control UI browser: the user says browser, dashboard, Control UI, or webchat; logs show
client=openclaw-control-uiormode=webchat. - Native mobile/node: the official app shows Connect, Scan QR, or setup code; logs/request metadata show a native client or
role=node.
A phone can be either client. Do not use openclaw qr or openclaw nodes status for a phone browser; those belong to native mobile/node pairing.
3. Observe the failing attempt
Run these through the locked target:
openclaw gateway status --deep
openclaw logs --follow --json
openclaw devices list
openclaw config get gateway.mode
openclaw config get gateway.bind
openclaw config get gateway.remote.url
openclaw config get gateway.auth.mode
openclaw config get gateway.auth.allowTailscale
openclaw config get gateway.tailscale.mode
Have the client retry once while logs are live. Correlate its client ID, mode, platform, address, auth result, device ID, user, and close code. Ignore other paired devices.
Interpret the first failed transition:
- No matching Gateway attempt: route, DNS, TLS, origin, or proxy problem.
token_missing,password_missing, mismatch, or Tailscale identity failure: auth failed before pairing, so an empty pending list is expected.pairing required: route and auth succeeded; approve the exact pending request.authenticated user connected/webchat connected: match the exact client; if followed by1006, preserve auth/pairing and inspect lifecycle, transport, proxy, or reconnect.1006without either application-level connected log does not prove auth/pairing; inspect the earlier handshake transition.
4. Prove the route
Choose one topology: same machine, LAN, tailnet, or public reverse proxy. Do not mix them.
- Browser Control UI needs HTTPS or localhost for browser device identity. A remote plain-HTTP Tailnet/LAN URL is not a valid substitute.
gateway.tailscale.mode=offmeans OpenClaw is not managing Serve/Funnel. It does not prove that Tailscale or an externally managed Serve route is absent.- When Tailscale is involved, inspect live state rather than inferring it from OpenClaw config:
tailscale status --json
tailscale serve status --json
Match the client's URL to the listener/proxy route reaching the locked Gateway.
5A. Control UI browser lane
Restore browser auth before looking for a pairing request:
- For token/password auth, enter the credential in Control UI settings. Never put permanent secrets in chat, logs, or URLs.
- Prefer
openclaw dashboardon the Gateway host for a one-time signed handoff. Use--no-openonly when the operator can retrieve that host's clipboard, and keep the host browser/clipboard outside agent tooling. Never capturedashboard --json: it can expose the handoff and shared credentials. Never relay, rewrite, or send a loopback handoff URL to a remote phone. - For Tailscale Serve, verify the live route and forwarded identity. Enable
gateway.auth.allowTailscaleonly for that intended trust boundary. Verified Tailscale Control UI auth with browser device identity can skip pairing.
After auth succeeds:
- Retry; never infer pairing from the pre-auth attempt. Token/password auth can reveal
pairing required; verified Tailscale identity can skip it. - If logs say
pairing required, re-list devices and approve the exact request ID. - A successful verified-Tailscale connection may correctly create no pending or paired-device row.
- After a repeated
1006, preserve auth/pairing and inspect reconnect evidence.
5B. Native mobile/node lane
Inspect the native route through the locked target without exposing the setup credential:
openclaw qr --json | jq '{gatewayUrl, gatewayUrls, auth, access, accessDowngraded, urlSource}'
For a CLI controlling a remote Gateway, add --remote before --json; it selects gateway.remote.url and remote credentials. If the redaction filter is unavailable, do not run raw QR JSON in agent-visible output.
Verify gatewayUrl and urlSource. The setup code is password-equivalent: have the operator copy it from Control UI → Devices → Pair device, or run openclaw qr --setup-code-only in a terminal outside agent tooling and paste it directly into the official app. Never relay it through agent/chat/tool output. Generate a fresh code after a URL/auth fix or expiry.
If the app reports pairing required:
openclaw devices list
openclaw devices approve --latest # preview only; exits without approval
openclaw devices approve <requestId>
openclaw nodes status
--latest only previews the current request; never treat it as approval. Re-list immediately before the exact-ID command because retries can supersede the request. Never approve by position, age, or similarity.
6. Correlate and finish
Before approval, match available request, device/public-key, client, mode/role, platform, address, user, and retry-time facts.
Declare success only after a new attempt made after the final change proves all applicable checks:
- the exact client reaches the locked Gateway;
- intended auth succeeds;
- the browser completes initial requests, or the native node appears in
openclaw nodes status; - approval used the exact request ID; and
- no immediate auth, pairing, or reconnect failure follows.
A QR/setup code, launched browser, empty pending list, approval, paired-device count, or Tailscale ping proves only one transition.
Report the diagnosis, chosen route/auth lane, exact-client evidence, and any remaining failed transition.
