fix(tailscale): bot computer commands time out on a tailnet host's 100.x address that the setup skill says works
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 52/100
- Issue type
- Bug
- Clarity
- Clearly specified
- Activity status
- Active
- Tech stack
- typescript
- Domain
- devtools, networking
Research direction
Read third_party/tailscale/skills/tailscale-setup/SKILL.md and the routing notes in third_party/tailscale/README.md, then trace the bot computer command route for 100.x destinations. Reproduce with the issue's curl loop while capturing traffic on the target host, and compare it with the MagicDNS route. Done when 100.x commands work as documented, or the skill accurately describes the supported route while preserving the successful by-name behavior.
Written by the indexing model from the issue text.
Description
Summary
The tailscale-setup skill says commands on Grok Bot's computer reach tailnet hosts by their name.tailnet.ts.net name or their 100.x address. On a connected personal tailnet, commands on the bot's computer timed out in both attempts that used a tailnet host's 100.x address, while both attempts that used the same host's MagicDNS name succeeded in the same run; a single by-address attempt in an earlier run also timed out. A packet capture on the target host saw no traffic from any by-address attempt. The by-name route works, and an MCP connector added by the same 100.x address also reached the host, so the failure is limited to bot computer commands that use a 100.x address.
Runtime or environment
- Grok Bot desktop 0.79.4 on macOS 27.2 (arm64), the latest release on Homebrew's
grok-botcask on 2026-11-11. - Tailscale plugin installed from the in-app catalog on 2026-11-11. The in-app plugin page shows no version, so the installed plugin version was not measured. The repository's default branch is at plugin version 1.0.6, commit 63912a91d52c43946f974f2aaf661334a6116a82.
- One personal tailnet, connected with one-click Connect from one Grok Bot account, so it was the default tailnet.
- Target: a plain-HTTP server on another device in the same tailnet, listening on all interfaces on port 3791. A
curlto its100.xaddress from its own host returned 200. - Not tested: company tailnets, HTTPS targets, non-HTTP TCP, devices shared in from another tailnet, other bots.
Steps to reproduce
- In Grok Bot desktop, open Connect apps, search for Tailscale, install the plugin, and connect a personal tailnet with the one-click Connect. Wait until the plugin shows the tailnet as Personal and Connected; a new Linux device for Grok Bot then shows as online in
tailscale statuson another tailnet machine. - On another device in that tailnet, run any plain-HTTP server on all interfaces (port 3791 here) and confirm it answers on its
100.xaddress locally. - About 90 seconds after the plugin shows Connected, as the bot's owner, send the bot one direct chat message (not a routine, Slack message, or teammate turn) asking it to run this on its own computer's terminal and return the raw output.
100.x.y.zis the server's Tailscale address and<host>.<tailnet>.ts.netits MagicDNS name:
for u in http://100.x.y.z:3791 http://<host>.<tailnet>.ts.net:3791; do for i in 1 2; do curl -sS -m 12 -A reach-test-bot-terminal -o /dev/null -w "$u http=%{http_code} ip=%{remote_ip}\n" "$u/.well-known/oauth-protected-resource/mcp"; echo "exit=$?"; done; done; env | grep -i proxy | sed 's/=.*/=<set>/'; getent hosts <host>.<tailnet>.ts.net || echo no-dns
- Optionally run
tailscale debug captureon the server host during step 3 to see which requests arrive from the Grok Bot device.
Expected behavior
The Bot computer section of the setup skill says: "On a connected tailnet, commands on Grok Bot's computer reach tailnet hosts by their name.tailnet.ts.net name or 100.x address, over HTTP(S) or TCP, through the same device." On a connected default tailnet, both URLs in step 3 should reach the server and return the same HTTP status.
Actual behavior
The bot's reply in the second run, with addresses and names replaced by placeholders:
curl: (28) Connection timed out after 12002 milliseconds
http://100.x.y.z:3791 http=000 ip=
exit=28
curl: (28) Connection timed out after 12002 milliseconds
http://100.x.y.z:3791 http=000 ip=
exit=28
http://<host>.<tailnet>.ts.net:3791 http=200 ip=198.18.128.1
exit=0
http://<host>.<tailnet>.ts.net:3791 http=200 ip=198.18.128.1
exit=0
SAND_HTTP_PROXY_NAME=<set>
SAND_EGRESS_TUNNEL_PROXY_ADDR=<set>
198.18.128.1 <host>.<tailnet>.ts.net
- Both by-address requests timed out (curl exit 28). Both by-name requests returned 200.
- The capture on the server showed the two by-name requests, with the marker User-Agent, arriving from the Grok Bot device's tailnet address and answered 200. It showed no packets at all for the two by-address attempts.
- In an earlier run about 25 minutes before, with a separate connection of the plugin and the message sent soon after connecting, a single by-address
curl -m 8from the bot's computer also timed out (exit 28). The capture showed no packets from the Grok Bot device. That run made no by-name attempt. The bot's computer reported no network interface with a100.address. - In the same sessions, an MCP connector added by the same
http://100.x.y.z:3791/mcpaddress reached the server from the Grok Bot device; its requests appeared in the capture.
Evidence
- Expected source:
third_party/tailscale/skills/tailscale-setup/SKILL.md, section "Bot computer", line 63, at the commit stated above; the same promise was added in 1.0.5 (third_party/tailscale/CHANGELOG.md: "reaching tailnet hosts by MagicDNS name or100.xaddress"). - Failure source: the bot's terminal output quoted under Actual behavior and the server-side
tailscale debug capturefrom the same two runs. - Evidence provenance: supplied
- Local verification: not-run
- Reproduction completeness: complete
The reporter ran both runs on 2026-11-11 through an agent session that drove Grok Bot desktop, and took the server-side capture. These runs were not repeated before filing because the test device and the plugin connection had been removed.
The by-name requests resolved the MagicDNS name to 198.18.128.1, and the bot's environment has proxy variables set. That suggests the bot computer's tailnet path intercepts traffic by host name, so a bare 100.x destination never enters it. This is an inference from the output, not an established cause.
Restoration check
On a connected default tailnet, run the step 3 command on the bot's computer. Today the two 100.x requests time out with http=000 and exit 28 while the two by-name requests return 200. The contract is restored when either the 100.x requests reach the server and return the same 200 as the by-name requests, with the server capture showing them arrive from the Grok Bot device, or the setup skill no longer says that commands on the bot's computer reach tailnet hosts by 100.x address and states what does work. The by-name requests must still return 200 in both cases.
Additional context
- The 1.0.6 changelog entry and the plugin description in
.cursor-plugin/plugin.jsonsay only that the bot computer reaches tailnet hosts "by name", while the skill's Bot computer section still says "or100.xaddress". - The skill's sentence "A short name or a
100.xaddress uses the default tailnet" in "Add MCP servers" is about MCP connectors and matches what was observed for the connector route. - The skill's general routing sentence ("A short name or a
100.xaddress goes to the member's default tailnet") and the same note inthird_party/tailscale/README.mdalso describe100.xaddresses without limiting them to MCP connectors; the skill's opening section then points readers to the Bot computer section for commands. - The skill's troubleshooting entry for a device shared in from another tailnet says it "can't be reached by name; use its
100.xaddress". If bot computer commands cannot use100.xaddresses, they may have no route to shared-in devices. That case was not tested.
References
- #1167 added the Bot computer section with the
100.xsentence in 1.0.5. - #1168 (1.0.6) describes the bot computer path in the changelog and plugin description as reaching hosts by name only, and left the skill's Bot computer sentence unchanged.
- #774 asked for non-MCP traffic from the bot's computer to use the plugin's connection; the Bot computer section documents that path.
Triage assessment
- Impact level: P2
- Assessment status: supported
- Impact basis: Following the setup skill, a command on the bot's computer that uses a tailnet host's
100.xaddress times out instead of reaching the host; both by-address attempts timed out while both by-name attempts to the same host succeeded in the same run, an earlier single by-address attempt also timed out, and a server-side capture showed no packets from any by-address attempt. The failure is bounded to by-address bot computer commands on the tested personal tailnet; MCP connectors and by-name commands work. - Workaround status: available
- Workaround basis: Using the host's full
name.tailnet.ts.netMagicDNS name from the bot's computer returned 200 in two of two attempts in the same session, with no setup change. Not tested: whether shared-in devices, which the skill says cannot be reached by name, have any working route from the bot's computer, and tailnets with MagicDNS turned off.
- Dominant language
- TypeScript
- Stars
- 18.2k
- Forks
- 2k
- Avg merge
- 3h 56m
- Merged PRs (30d)
- 96
Getting set up
This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from cursor/plugins
-
pstack: interrogate asks the user, which worker.md forbids for delegates that reach itPossibly taken A pull request linked to this issue is open or already merged. Open
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
cursor/plugins#1272 · 1 comment ·
Maintainers usually reply within 1 day
-
poteto-mode: multi-phase-plan Merge placeholder still appends every PR to a base-branch stackPossibly taken @mangeshraut712 claimed this 15 days ago. Open
Difficulty 1/5 Under an hour Newbie friendliness 78/100
Maintainers usually reply within 1 day
-
typescript-best-practices: schema example uses z.string().uuid(), deprecated in Zod 4Possibly taken @mangeshraut712 claimed this 14 days ago. Open
Difficulty 1/5 Under an hour Newbie friendliness 88/100
cursor/plugins#1165 · 1 comment ·
Maintainers usually reply within 1 day
-
statechart: relative-symlink launcher test fails on macOS because /var is a symlinkPossibly taken A pull request linked to this issue is open or already merged. Open
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
cursor/plugins#1164 · 2 comments ·
Maintainers usually reply within 1 day
-
pstack docs/harness/claude-code.md: transcript slug replaces every non-alphanumeric character, not only / and .Possibly taken @isTuhin claimed this 9 days ago. Open
Difficulty 1/5 1-3 hours Newbie friendliness 82/100
cursor/plugins#1155 · 2 comments ·
Maintainers usually reply within 1 day
Similar issues
-
[UI] Guard webApis against undefined window in SSR environmentsPossibly taken @madhavzanwar claimed this today. Open
Difficulty 2/5 1-3 hours Newbie friendliness 62/100
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 62/100
headroomlabs-ai/headroom#4992 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 65/100
paperclipai/paperclip#22575 ·
Maintainers usually reply within 1 day
-
community first-timers-only good first issue hacktoberfest help wanted low hanging fruit up-for-grabs
Difficulty 1/5 Under an hour Newbie friendliness 82/100
lingdojo/kana-dojo#34502 · 1 comment · 5 reactions ·
Maintainers usually reply within 1 day
-
bug frontend good first issue help wanted
Difficulty 2/5 1-3 hours Newbie friendliness 86/100
nikhilb2/simple_invoicing#935 ·