What
An MCP task is a long-running job started through an MCP server (a connected tool provider). In cloud sessions on Anthropic's infrastructure (CLAUDE_CODE_REMOTE true, environment kind unset or anthropic_cloud), a running task used to keep the cloud container, the machine the session runs on, up without limit.
- After a number of hours a running
mcp_taskstops keeping the container up. The default is 168 hours (7 days); the server-controlled valuetengu_curious_rivercan override it with a number or numeric text. - At the limit the task is marked
holdCapped, keeps being checked, and a message says it "has reached the … h limit on keeping its cloud container up". - The cap is also applied when a session is restored, based on when the task started.
- A task marked
holdCappedis skipped by one completion notice path. - Separately, renaming a job in the agent list now records whether it is a cloud job.
Why
A forgotten long-running task no longer keeps a cloud session's machine alive indefinitely, but once the limit passes the container may shut down while the task is still running.
tengu_curious_river Not enough to sayNothing here resolved what this flag was doing on this version, so nothing here should be read as on or off.
This account: no value returned · anonymous baseline: no value returned · compiled default in v2.1.305: not a boolean we can read
These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.
Read once, for one account on one subscription tier, against v2.1.305. It isn't a statement about your account. What a flag value here can and cannot tell you
A reading is one sample. Claude Code evaluates its flags remotely, so no client sees the targeting rule behind a value and this says nothing about your account.
A reading is one sample. Claude Code evaluates its flags remotely, so no client sees the targeting rule behind a value and this says nothing about your account.
A reading is one sample. Claude Code evaluates its flags remotely, so no client sees the targeting rule behind a value and this says nothing about your account.
A reading is one sample. Claude Code evaluates its flags remotely, so no client sees the targeting rule behind a value and this says nothing about your account.
A reading is one sample. Claude Code evaluates its flags remotely, so no client sees the targeting rule behind a value and this says nothing about your account.
New in this build: tengu_curious_river