Summary
Once a tab has been driven (screenshot / navigate / javascript_execute), later calls against that same tab intermittently fail with either "Another debugger is already attached to the tab with id: N" or a timeout ("TaskWindow extension did not answer within 10s/30s"). A fresh tabs_create works once, then that tab shows the same behaviour after its first navigate. Over a long session this forces one new tab per interaction and leaves the tab strip full of one-tab groups.
Environment
- Extension 0.2.4, macOS (Darwin 25.5.0), Chrome
- Called over MCP (
mcp__taskwindow__*) from an MCP client; single sessionToken for the whole session
- Target pages:
http://localhost:5173/* (Vite dev server) and https://app.attio.com/*
Reproduction
tabs_create → tab A
computer { action: "screenshot" } on A → OK
navigate A to another URL on the same origin → OK
computer { action: "screenshot" } on A → "Another debugger is already attached to the tab with id: A"
tabs_create → tab B; screenshot OK; navigate B; screenshot → same error on B
Also observed, same session:
tabs_close on a tab in this state times out (10s) but the tab sometimes closes anyway
tabs_list itself timing out (10s)
- one "TaskWindow extension is not connected" between otherwise-working calls, with the options page reporting Connected
Expected
A tab the session created should stay drivable across navigations; tabs_close should either close it or return an error, not time out.
Guess at the cause
Looks like the CDP Debugger.attach for a tab is not released (or is re-acquired) around navigate, so the second attach sees the first still held. The timeouts on tabs_list / tabs_close suggest the same held attachment blocks the message loop.
Workaround
Create a new tab for each interaction. Works, but every tabs_create with a new task name makes a new tab group, which is how the strip ends up as seven one-tab groups.
Summary
Once a tab has been driven (screenshot / navigate / javascript_execute), later calls against that same tab intermittently fail with either "Another debugger is already attached to the tab with id: N" or a timeout ("TaskWindow extension did not answer within 10s/30s"). A fresh
tabs_createworks once, then that tab shows the same behaviour after its first navigate. Over a long session this forces one new tab per interaction and leaves the tab strip full of one-tab groups.Environment
mcp__taskwindow__*) from an MCP client; singlesessionTokenfor the whole sessionhttp://localhost:5173/*(Vite dev server) andhttps://app.attio.com/*Reproduction
tabs_create→ tab Acomputer { action: "screenshot" }on A → OKnavigateA to another URL on the same origin → OKcomputer { action: "screenshot" }on A → "Another debugger is already attached to the tab with id: A"tabs_create→ tab B; screenshot OK; navigate B; screenshot → same error on BAlso observed, same session:
tabs_closeon a tab in this state times out (10s) but the tab sometimes closes anywaytabs_listitself timing out (10s)Expected
A tab the session created should stay drivable across navigations;
tabs_closeshould either close it or return an error, not time out.Guess at the cause
Looks like the CDP
Debugger.attachfor a tab is not released (or is re-acquired) aroundnavigate, so the second attach sees the first still held. The timeouts ontabs_list/tabs_closesuggest the same held attachment blocks the message loop.Workaround
Create a new tab for each interaction. Works, but every
tabs_createwith a new task name makes a new tab group, which is how the strip ends up as seven one-tab groups.