Anthropic released Claude Code 2.1.319 on November 6, moving MCP tool calls into the background
Claude Code 2.1.319 immediately moves MCP tool calls that run as tasks into the background and notifies the model on completion, adds a permissions.denyInsteadOfPrompting setting, and adjusts auto mode behavior behind LLM gateways.
On November 6, Anthropic released Claude Code 2.1.319. The central change is how MCP tasks execute: when an MCP server runs a tool call as a task, Claude Code immediately moves it into the background and tells the model the result when it finishes, running without a feature flag, with availability on Bedrock, Vertex, Foundry, LLM gateways and in environments with telemetry disabled, and an environment variable to turn it off. Version 2.1.318 also moved background subagent shell commands longer than three minutes into the background to avoid prompt cache expiry during execution.
The new version adds a permissions.denyInsteadOfPrompting setting, so tool calls that would wait on a permission prompt are denied instead and the turn continues, applying to terminal sessions, subagents and teammates, which prevents unattended runs from stalling on unanswered prompts. The same release fixed multiple Bash permission bypass issues, including a path where a crafted environment variable prefix could hide part of a command from permission checks.
Auto mode behavior was also adjusted. When telemetry or feature flags are off, including behind LLM gateways, calls the server cannot check are asked about or denied with no fallback to the built-in classifier, and a new environment variable explicitly disables the server-side check. In addition, a fix means the command line now exits with 1 instead of 0 when auto mode refuses an action but no safety check is available, for both text and JSON output.