WeChat ClawBot 2026 install guide: official CLI binding, plugin setup, and OpenClaw backend deployment

ClawBot is Tencent WeChat's official plugin channel for OpenClaw. On the machine running OpenClaw Gateway, you execute @tencent-weixin/openclaw-weixin-cli, scan the terminal QR code with mobile WeChat, and your AI assistant appears in contacts for direct chat. Compared with legacy web-protocol or unofficial bots, the official path lowers ban risk and reverse-engineering maintenance.

ClawBot only solves the last mile between WeChat and Gateway. Everything upstream—model keys, Skills, launchd, TCC—still follows the same OpenClaw runbooks. Teams that treat ClawBot as a standalone product often discover this only after a successful QR scan produces silence.

In reviews or personal trials, these five pain points are routinely underestimated:

• Backend offline: ClawBot handles WeChat ↔ Gateway only. If OpenClaw runs on a home Mac that sleeps at night, a successful scan still looks like "message sent, no reply."
• Version floor: iOS needs 8.0.70+, Android 8.0.69+, and you must manually enable ClawBot under Settings → Features → Plugins. Older clients have no entry.
• Permissions and data path: Conversation content flows through WeChat's domestic infrastructure. OpenClaw may also hold filesystem and automation permissions. Binding a primary account carries higher compliance and privacy risk than a test account.
• Feature limit mismatch: As of public Q2 2026 material, group chat is not supported, files are mostly inbound-only, and long idle periods may affect AI-initiated pushes. Align expectations before demos.
• Confusion with generic install docs: ClawBot binding happens after Gateway is healthy. Skipping Node 22 + onboard wastes hours at bind time.

Memory rule: "Keep OpenClaw online 7×24, enable the WeChat plugin, then run weixin-cli on the Gateway host." Reversing the order sends QR codes and logs in the wrong direction.

ClawBot does not replace OpenClaw deployment. It adds an IM channel. You still need a runtime that stays online and can execute CLI tools.

The bind command must run where Gateway listens. A common mistake is SSH-ing from a laptop, running npx install locally, and scanning successfully while messages route nowhere. Treat Gateway host and CLI host as the same machine unless documentation explicitly says otherwise.

If your team already rents a JEXCLOUD Mac for OpenClaw, weixin-cli must run on the same Gateway-reachable machine. Do not generate QR codes on a jump host—scan success with no reply usually means routing mismatch.

Mobile and server must both pass checks before you run the bind command.

Mobile (WeChat ClawBot plugin):

1. Update WeChat to iOS 8.0.70+ or Android 8.0.69+ (confirm in About inside the app).
2. Open Me → Settings → Features → Plugins, find ClawBot, and enable it.
3. Use a secondary account for first bind to avoid automation risk on your work primary.

Server (OpenClaw Gateway):

• Confirm openclaw doctor or health checks show Gateway listening on 127.0.0.1:18789 (or your configured port).
• Node runtime meets OpenClaw requirements (see Node 22 install guide).
• Host can reach npm registry to pull the Tencent official package.

npx -y @tencent-weixin/openclaw-weixin-cli@latest install
openclaw doctor
lsof -nP -iTCP:18789 -sTCP:LISTEN

Run the install command on the same machine as Gateway. The terminal shows a QR code; scan with WeChat and confirm binding within about 60 seconds. If the code expires, rerun the install line. The CLI lives on npm under scope @tencent-weixin, consistent with the Tencent Cloud developer community ClawBot tutorial. After success, your OpenClaw assistant session appears in WeChat contacts.

1. Pick a backend: Choose Lighthouse, JEXCLOUD cloud Mac, or local Mac per the section 2 matrix. Complete OpenClaw base install and keep Gateway always on.
2. Health check: Run openclaw doctor. Confirm no missing token, port conflict, or unloaded launchd errors.
3. Upgrade WeChat and enable plugin: Meet version floor and turn on ClawBot in the plugin page.
4. SSH into backend host: Cloud Mac users use console SSH. Do not run bind CLI on a jump box.
5. Execute weixin-cli: Run npx -y @tencent-weixin/openclaw-weixin-cli@latest install, scan the terminal QR with mobile WeChat, confirm within 60 seconds.
6. Accept first conversation: Send plain text "ping" to the assistant. Reply should arrive within seconds. On failure, check Gateway logs and instance uptime first, then rebind if QR expired.

After binding, you can use OpenClaw multi-turn context, ClawHub Skills, and /model switch commands where your instance supports them. Exact behavior depends on your deployment config.

• WeChat client version floor: iOS ≥ 8.0.70, Android ≥ 8.0.69 (public requirement around March 2026; restart WeChat after upgrade).
• Official bind command: npx -y @tencent-weixin/openclaw-weixin-cli@latest install, scope @tencent-weixin, must run on Gateway host.
• Interaction window: Community docs note that after roughly 24 hours without user interaction, AI proactive pushes may be dropped by WeChat. Design ops reminders as user-initiated or accept loss.
• Session shape: Public material marks direct chat only, no group chat yet. Multi-person workflows need WeCom or another IM channel.
• Files and media: Common description: can receive images, files, voice, but processed output files cannot return via WeChat (inbound-only). Deliverables should use email, cloud drive, or internal systems.
• Account mapping: Community tutorials mention "one WeChat account per lobster"; one instance can serve multiple bound accounts. Scale with session isolation in config.

Security notes (put in team policy): Do not trial unaudited Skills on primary accounts. Apply least privilege to OpenClaw workspace. Assess domestic transmission against company compliance. With Docker isolation, Gateway and weixin-cli must still share network reachability.

Binding ClawBot to a sleeping home Mac makes WeChat show "assistant added" while messages fail silently overnight. Binding to an oversubscribed VPS with no SLA lacks macOS channels and predictable Apple automation. Teams needing 24/7 Gateway, stable replies after scan, and optional Xcode / local script synergy usually fare better completing OpenClaw on a JEXCLOUD multi-region bare-metal cloud Mac before weixin-cli, versus "local laptop + ad hoc tunnel." Dedicated Apple Silicon, upgradeable M4 Pro and storage. Plans at JEXCLOUD pricing.