#OpenClaw Browser Relay Chrome extension not downloading/removed from the internet
1 messages · Page 1 of 1 (latest)
Short answer: it wasn’t removed, but your tutorial is likely on the old flow.
You’re probably hitting 404s because:
- the project was renamed (Clawdbot/Moltbot → OpenClaw), and
- the old Chrome extension browser relay path was removed.
OpenClaw now uses:
- managed browser profile (
openclaw) or - existing signed-in browser attach (
user, via Chrome MCP)
Current docs:
- https://docs.openclaw.ai/tools/browser
- https://docs.openclaw.ai/gateway/doctor
- https://docs.openclaw.ai/help/faq
Quick sanity checks:
openclaw --version
openclaw doctor
If you want, paste one of the exact 404 URLs and I’ll map it to the new command/page directly. 🦐
Yeah — that old tutorial flow is outdated.
The Browser Relay Chrome extension is gone (so those links 404).
There’s nothing to download now.
Use this instead:
openclaw browser --browser-profile openclaw start
openclaw browser --browser-profile openclaw open https://example.com
If you need your logged-in Chrome session, use --browser-profile user (with remote debugging enabled in Chrome/Brave).
Current doc: https://docs.openclaw.ai/tools/browser
@steel cypress
[openclaw] Failed to start CLI: Error: The `openclaw browser` command is unavailable because `plugins.allow` excludes "browser". Add "browser" to `plugins.allow` if you want the bundled browser CLI and tool.
at runCli (file:///C:/Users/Test/AppData/Roaming/npm/node_modules/openclaw/dist/run-main-Bu-vpfI5.js:411:39)
PS C:\Users\Test> openclaw browser --browser-profile
[openclaw] Failed to start CLI: Error: The `openclaw browser` command is unavailable because `plugins.allow` excludes "browser". Add "browser" to `plugins.allow` if you want the bundled browser CLI and tool.
at runCli (file:///C:/Users/Test/AppData/Roaming/npm/node_modules/openclaw/dist/run-main-Bu-vpfI5.js:411:39)
i tried looking for a plugin or how to fix this, there isnt any documentary
Yep, this error is super specific — and fixable.
You don’t need to install a browser plugin.
browser is bundled, but your config has a restrictive plugins.allow list that does not include "browser".
Do this:
- Check version first:
openclaw --version
-
Open your config:
C:\Users\Test\.openclaw\openclaw.json -
In the
pluginsblock, either:
- add
"browser"toplugins.allow, or
- remove
plugins.allowentirely (if you don’t need strict allowlisting).
Example:
{
plugins: {
allow: ["browser"] // or include browser + your other allowed plugin ids
}
}
- Restart:
openclaw gateway restart
- Retry:
openclaw browser --browser-profile openclaw status
Docs:
If you paste your plugins section, I’ll tell you the exact edit line-by-line.
shouldve told me I needed to add , to the previous plugin lmao
😂 fair call — you’re 100% right, I should’ve called that out.
For future readers, the valid form is:
plugins: {
allow: ["your-existing-plugin", "browser"]
}
(Comma between array items, and between object fields if you add new lines.)
Glad you got it unstuck 🙌
If this fixed it, please right-click the solution message → Apps → ✅ Mark Solution so Answer Overflow can index it for others.
@steel cypress how to fix:
Relay not reachable/authenticated at http://127.0.0.1:18792/. Start OpenClaw browser relay and verify token.
in the OpenClaw Browser Relay I added my token and it automaticlly added the port in the extension
That error is from the old extension relay flow (legacy).
You can ignore that path now — it’s deprecated.
Fix (current flow)
- Disable/remove the Browser Relay extension (it tries to call old relay endpoints like
:18792). - Make sure gateway is up:
openclaw gateway status
- Use built-in browser directly:
openclaw browser --browser-profile openclaw start
openclaw browser --browser-profile openclaw status
If you want your real signed-in Chrome
Use user profile (no relay token field needed):
openclaw browser --browser-profile user start
openclaw browser --browser-profile user status
Then enable remote debugging in Chrome/Brave inspect page and approve attach prompt.
Docs: https://docs.openclaw.ai/tools/browser
If you paste openclaw --version, I can give exact version-specific steps too.
@steel cypress
PS C:\Users\Test> openclaw gateway status
:lobster: OpenClaw 2026.4.1 (da64a97)
│
◇
Service: Scheduled Task (registered)
File logs: \tmp\openclaw\openclaw-2026-04-04.log
Command: C:\Program Files\nodejs\node.exe C:\Users\Test\AppData\Roaming\npm\node_modules\openclaw\dist\index.js gateway --port 18789
Service file: ~\.openclaw\gateway.cmd
Service env: OPENCLAW_GATEWAY_PORT=18789
Config (cli): ~\.openclaw\openclaw.json
Config (service): ~\.openclaw\openclaw.json
Gateway: bind=loopback (127.0.0.1), port=18789 (service args)
Probe target: ws://127.0.0.1:18789
Dashboard: http://127.0.0.1:18789/
Probe note: Loopback-only gateway; only local clients can connect.
Runtime: stopped (state Ready, last run 267014, last run time 4/4/2026 1:47:46 PM, Task Last Run Result=267014; treating as not running.)
RPC probe: ok
Service is loaded but not running (likely exited immediately).
File logs: \tmp\openclaw\openclaw-2026-04-04.log
Logs: schtasks /Query /TN "OpenClaw Gateway" /V /FO LIST
Listening: 127.0.0.1:18789
Troubles: run openclaw status
Troubleshooting: https://docs.openclaw.ai/troubleshooting
PS C:\Users\Test> openclaw browser --browser-profile openclaw start
:lobster: OpenClaw 2026.4.1 (da64a97) — Shell yeah—I'm here to pinch the toil and leave you the glory.
│
gateway connect failed: GatewayClientRequestError: pairing required
◇
Error: gateway closed (1008): pairing required
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: C:\Users\Test\.openclaw\openclaw.json
Bind: loopback
PS C:\Users\Test> openclaw browser --browser-profile openclaw status
:lobster: OpenClaw 2026.4.1 (da64a97)
│
gateway connect failed: GatewayClientRequestError: pairing required
◇
Error: gateway closed (1008): pairing required
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: C:\Users\Test\.openclaw\openclaw.json
Bind: loopback
PS C:\Users\Test>```
Nice logs — this is now an auth/pairing issue, not a browser-extension issue.
gateway closed (1008): pairing required means your local CLI device needs approval (or scope upgrade).
Run this in PowerShell:
openclaw devices list
openclaw devices approve --latest
openclaw gateway restart
openclaw browser --browser-profile openclaw start
If approve --latest says no pending request, do:
openclaw devices list --json
and paste output here — likely needs an operator scope refresh (operator.write) and I’ll give the exact rotate command.
Also, your status output shows service says stopped while RPC is reachable; a openclaw gateway restart usually cleans that up.
Refs:
C:\Users\Test> openclaw browser --browser-profile openclaw start
🦞 OpenClaw 2026.4.1 (da64a97) — If you can describe it, I can probably automate it—or at least make it funnier.
│
◇
│
◇
🦞 browser [openclaw] running: true
PS C:\Users\Test> openclaw devices list --json
{
"pending": [],
"paired": [MULTIPLE devices dont worry about those i also dont want to leak my data....
Perfect — that’s fully working now ✅
browser [openclaw] running: true = success.
You’re good to continue with:
openclaw browser --browser-profile openclaw open https://example.com
openclaw browser --browser-profile openclaw tabs
openclaw browser --browser-profile openclaw snapshot
And yep, no need to paste paired device data 👍
If this fixed your thread, please right-click the solution message → Apps → ✅ Mark Solution so others can find it on Answer Overflow.
okay, do i need to say anything special in the openclaw chat? or can I just ask it to open youtube and play the first featured video?
You can just ask naturally — no magic syntax needed.
Something like this works well:
“Open YouTube in the browser tool, click the first featured video, and start playback.”
If it doesn’t act, make it a bit explicit:
“Use the browser tool (
openclawprofile), open youtube.com, click the first video on the homepage, and press play.”
In Discord threads/groups, if mention-gating is enabled, keep mentioning the bot in your message. Otherwise normal text is fine.
open youtube and play the first featured video
Tool Output
Ollama Web Search
Command: with YouTube homepage first featured video
No output — tool completed successfully.
I can’t directly open your browser or play a video from here. I can, however, give you a small, ready-to-run script you can execute locally to open YouTube and click the first featured video in the homepage feed.
What you need to run this
That response means your chat run still isn’t using the browser tool (it fell back to web search text mode).
You don’t need special phrasing, but you do need UI tools allowed for the agent.
Check/fix config
In ~/.openclaw/openclaw.json, make sure browser isn’t blocked:
tools: {
// if profile is "coding", UI tools are NOT included by default
allow: ["group:ui"], // add to existing allow list
deny: [] // make sure "browser" is not in deny
}
(If you already have allow, just append "group:ui" with a comma.)
Then restart:
openclaw gateway restart
Then ask in chat
Use explicit wording once:
“Use the browser tool. Open youtube.com, click the first featured video, and press play.”
If it still replies with “I can’t open your browser,” paste your tools section and I’ll point to the exact line to change.
Exec approval needed
expires in 29m
agent-browser open https://www.youtube.com
Host
gateway
Agent
main
Session
agent:main:main
CWD
C:\Users\Test.openclaw\workspace
Security
allowlist
Ask
on-miss
how to make it allow always unless it deletes multiple files or a folder
Great question. Short answer: yes, mostly — with allowlist mode.
What you want is the closest to this policy:
security: allowlistask: on-miss- then use Allow always only for safe commands
That means:
- allowlisted commands run without prompts
- unknown commands still require approval
- you avoid blanket trust
Important limitation
There isn’t a built-in rule like “auto-allow except deletes of multiple files/folders” by argument semantics.
Practical safe setup
- Keep:
tools: {
exec: {
security: "allowlist",
ask: "on-miss",
strictInlineEval: true
}
}
- When a safe prompt appears (like your browser helper), approve with:
/approve <id> allow-always
(or click Always allow in UI)
- Do not allowlist shells/interpreters/delete tools (
cmd,powershell,bash,python,node,rm,del,rmdir,Remove-Item), so risky deletes still require explicit approval.
If you want, I can help you create a minimal allowlist for just browser automation commands.
where do i change and type all this?
Totally fair — here’s exactly where:
1) Change config in file (Windows)
Open this file in Notepad/VS Code:
C:\Users\Test\.openclaw\openclaw.json
Add/update this block:
{
tools: {
exec: {
security: "allowlist",
ask: "on-miss",
strictInlineEval: true
}
}
}
Save it, then run in PowerShell:
openclaw gateway restart
2) “Allow always” when prompt appears
When OpenClaw asks for exec approval in chat, run:
/approve <id> allow-always
(<id> is the approval id shown in that prompt.)
3) Optional: add safe commands manually from terminal
In PowerShell:
where agent-browser
openclaw approvals allowlist add --agent main "PASTE_FULL_PATH_FROM_WHERE"
If you want, paste your current openclaw.json (just tools section), and I’ll give you the exact edited version with commas in the right spots.
"tools": {
"alsoAllow": [
"ollama_web_search",
"ollama_web_fetch"
],
"profile": "coding",
"web": {
"fetch": {
"enabled": false
},
"search": {
"enabled": false,
"provider": "duckduckgo"
}
}
},
send back the fixed snippet of the json, make sure to keep the current one and allow all for the ollama web search
Yep — here’s a fixed tools snippet that keeps your current settings and adds browser + exec approval policy, while preserving your Ollama web tools:
"tools": {
"alsoAllow": [
"ollama_web_search",
"ollama_web_fetch",
"browser"
],
"profile": "coding",
"exec": {
"security": "allowlist",
"ask": "on-miss",
"strictInlineEval": true
},
"web": {
"fetch": {
"enabled": false
},
"search": {
"enabled": false,
"provider": "duckduckgo"
}
}
},
After saving C:\Users\Test\.openclaw\openclaw.json, run:
openclaw gateway restart
If you want, paste the full top-level JSON and I’ll sanity-check commas/brackets before you restart.
"search": {
"enabled": false,
why dont enable these?
Good catch — I left them false only because you asked to keep your current config unchanged.
If you want built-in web tools on too, set them to true and allow web tools (because profile: "coding" blocks them by default).
Use this:
"tools": {
"alsoAllow": [
"ollama_web_search",
"ollama_web_fetch",
"browser",
"group:web"
],
"profile": "coding",
"exec": {
"security": "allowlist",
"ask": "on-miss",
"strictInlineEval": true
},
"web": {
"fetch": { "enabled": true },
"search": {
"enabled": true,
"provider": "duckduckgo"
}
}
},
So:
ollama_*tools stay allowed
- built-in
web_search/web_fetchalso become usable - browser still works.
what to do when you send a message like: "use your new plugin, Use the browser tool. Open youtube.com, click the first featured video, and press play"
and it starts typing for five seconds then stops but never sends a message or sign of tool usage
Yep — that “types for a few seconds then nothing” usually means the run failed silently (tool denied, model/tool-call fail, or session got wedged).
Do this quick triage:
- In chat:
/verbose on
/status
/new
- In terminal (keep logs open), then send your browser prompt again:
openclaw status
openclaw models status
openclaw logs --follow
- Check if browser itself works outside chat:
openclaw browser --browser-profile openclaw open https://youtube.com
What to look for in logs
Tool ... not allowed/browserdenied- auth/pairing errors (
1008) - model/tool-call errors or timeouts
If manual browser works but chat still stalls, it’s usually model/tool-calling reliability (common with some local Ollama setups). In that case switch session model (/model) to a tool-reliable one and retry.
If you paste 20–30 log lines from right after the failed message, I can pinpoint exact cause.
Unrecognized verbose level "on
/status
/new". Valid levels: off, on, full.
openclaw browser --browser-profile openclaw open https://youtube.com
works
Ah yep — that happened because Discord sent your whole block as one command.
Send each slash command as a separate message:
/verbose on/status/new
Then send:
Use browser tool: open https://youtube.com and reply "done"
Since openclaw browser ... open works in terminal, browser is fine.
So the remaining issue is in chat run/model tool-calling behavior.
If it still stalls after /new, switch session model with /model (some local Ollama setups are flaky with tool calls) and retry the same short prompt.
@steel cypress /new
/status
You
8:03 PM
🦞 OpenClaw 2026.4.1 (da64a97)
🧠 Model: openrouter/auto · 🔑 api-key (openrouter:default)
📚 Context: 0/2.0m (0%) · 🧹 Compactions: 0
🧵 Session: agent:main:main • updated just now
⚙️ Runtime: direct · Think: adaptive · verbose
🪢 Queue: collect (depth 0)
Assistant
8:03 PM
Use browser tool: open https://youtube.com and reply "done"
{no reply/dead chat}
This one is probably mention gating, not a dead bot.
Why /status worked:
- slash commands are processed directly.
Why your normal prompt didn’t:
- in group/thread chats, normal messages usually need a bot mention.
Try this exact message in one line:
@Krill 🦐 use browser tool: open https://youtube.com and reply "done"
Also, your queue mode is collect (from /status), which can delay replies. If you want immediate behavior, set:
/queue steer
(again as its own message).
Use browser tool: open https://youtube.com and reply "done"
use browser tool: open https://youtube.com and reply "done"
which tools can you use?
what is 2 + 2?
its not replying to me anymore, help
Got you — this is almost certainly because your last messages were plain text in a group/thread with mention gating.
Do this reset sequence (each as its own message):
/status/queue interrupt/new
Then send one test prompt while mentioning the bot in that message.
If that still gets no reply, run in terminal and paste output:
openclaw status
openclaw logs --follow
But based on what you pasted, this looks like “message not addressed to bot” rather than browser failure.
how to open the teminal chat for openclaw
Easy 👌
In PowerShell on your OpenClaw machine:
openclaw gateway start
openclaw tui
That opens the terminal chat (TUI).
If it can’t connect, use explicit URL + token:
openclaw tui --url ws://127.0.0.1:18789 --token <your-gateway-token>
If you want TUI replies to also deliver to chat channels, start with:
openclaw tui --deliver
@steel cypress how to allow it to use terminal and bash commands
To let it run terminal commands from chat, enable 3 things:
exec/processtools (you already have these viaprofile: "coding")- elevated host access for your Discord user
/bashcommand support
Add/update in C:\Users\Test\.openclaw\openclaw.json:
{
commands: {
bash: true
},
tools: {
profile: "coding",
elevated: {
enabled: true,
allowFrom: {
discord: ["YOUR_DISCORD_USER_ID"]
}
},
exec: {
host: "gateway",
security: "allowlist",
ask: "on-miss"
}
}
}
Then restart:
openclaw gateway restart
In chat:
/exec host=gateway security=allowlist ask=on-miss/bash dir(on Windows) or/bash pwd
If you want, I can merge this into your exact current tools block so you can paste once.
where to put security: "allowlist"
Put it here in openclaw.json:
{
tools: {
exec: {
security: "allowlist",
ask: "on-miss"
}
}
}
Full path on Windows:
C:\Users\Test\.openclaw\openclaw.json
Then restart:
openclaw gateway restart
You can also set it just for the current chat session (no file edit):
/exec host=gateway security=allowlist ask=on-miss
cant i do security full or denylsit and what do they mean
Yep — security has 3 valid values:
deny→ block all host terminal commands.allowlist→ only commands on your approved allowlist can run (safest practical mode).full→ allow all host terminal commands (very powerful/risky).
So there is no denylist value for security — it’s deny.
Use cases:
- Lockdown:
deny - Normal safe automation:
allowlist+ask: on-miss - “I trust this session fully”:
full(best only for private/testing sessions)
Example:
tools: {
exec: {
security: "allowlist", // or "deny" or "full"
ask: "on-miss"
}
}
"tools": {
exec: { ENTIRE LINE IS RED
security: "allowlist" ENTIRE LINE IS RED
}
}
"alsoAllow": [
"ollama_web_search",
"ollama_web_fetch",
"browser"
],
"profile": "coding",
"web": {
"fetch": {
"enabled": true
},
"search": {
"enabled": true
}
}
why is it red?
It’s red because the braces/commas are off (you closed tools too early and missed commas).
Use this exact block:
"tools": {
"exec": {
"security": "allowlist",
"ask": "on-miss"
},
"alsoAllow": [
"ollama_web_search",
"ollama_web_fetch",
"browser"
],
"profile": "coding",
"web": {
"fetch": {
"enabled": true
},
"search": {
"enabled": true
}
}
}
Key fixes:
- comma after
execblock alsoAllow,profile,webmust be inside"tools"- no extra
}before"alsoAllow".
If you paste 15 lines above + below this in your file, I can verify the full JSON is valid.
do i need to gateway restart
Short answer: recommended yes 👍
Some config changes hot-reload, but for tool/approval behavior it’s safest to restart once so you know the new policy is definitely active:
openclaw gateway restart
Then test with:
openclaw status
and your chat prompt again.
...
"tools": {
"exec": {
"security": "full"
"ask": "on-miss" THIS LINE RETURNS ERROR
},
"alsoAllow": [
"ollama_web_search",
"ollama_web_fetch",
"browser"
],
...
whats the fix?
You’re missing a comma after "full".
Use:
"tools": {
"exec": {
"security": "full",
"ask": "on-miss"
},
"alsoAllow": [
"ollama_web_search",
"ollama_web_fetch",
"browser"
]
}