`claude agents` Gains Full Configuration Flag Support for Background Sessions
Claude CodeClaude Code v2.1.142 adds eight new flags to the claude agents command — including --add-dir, --settings, --mcp-config, --plugin-dir, --permission-mode, --model, --effort, and --dangerously-skip-permissions — giving developers full control over how background sessions are configured when dispatched. Previously, background agents inherited defaults without the ability to specify custom tooling or model settings at dispatch time. These flags make claude agents a first-class entrypoint for configuring multi-agent workflows.
Sources & Mentions
5 external resources covering this update
Manage multiple agents with agent view
Anthropic Docs
Thoughts on Claude Code 2.1.139 Agent View and Background Sessions
Hacker News
Releases · anthropics/claude-code
GitHub
How to Use Claude Code's /bg Command to Run Background Agent Sessions
MindStudio
Claude Code: Agent View, Goal Command, and Background Sessions Update
Geeky Gadgets
claude agents Now Accepts Full Configuration at Dispatch Time
Claude Code v2.1.142 significantly expands the claude agents command with eight new configuration flags, bringing its configurability on par with the main claude CLI. Developers launching background or parallel sessions can now specify exactly which settings, MCP servers, plugins, directories, permission mode, model, and effort level each session should use — without relying on ambient defaults or environment variables.
New Flags at a Glance
The complete set of new flags added in v2.1.142:
--add-dir— Mount additional directories into the agent session's working context--settings— Point the session at a custom settings file rather than the default~/.claude/settings.json--mcp-config— Load a specific MCP configuration for the session's tool connections--plugin-dir— Specify a plugin directory to load for the dispatched session--permission-mode— Set the session's permission mode (e.g.plan,bypassPermissions)--model— Override the model for this session--effort— Set the effort level (e.g.high,xhigh) for this dispatch--dangerously-skip-permissions— Bypass permission prompts for fully automated runs
Why This Matters for Multi-Agent Workflows
Before v2.1.142, dispatching a background agent meant accepting whatever defaults the system provided. Teams managing complex multi-agent setups — where different agents might need different MCP connections, different permission modes, or different models — had no clean way to express those distinctions at launch time. With these flags, each dispatch call can be fully self-contained.
This is especially useful for CI/CD-style pipelines where agents are dispatched programmatically: a code review agent might need a different model and stricter permission mode than a refactoring agent, and both can now be configured independently in the same claude agents invocation flow.