Skip to main content
Claude Code comes with Mains. Sign in during onboarding or open Settings → Claude to inspect your account and runtime. Claude supports Code, Work, Chat, and Atlas.

​
Settings and conversation controls

Settings includes the default permission mode, output style, and structured output. Choose a model, thinking, effort, and supported fast options in the composer. Conversation settings stay with each chat. Opening an older chat does not replace the defaults for new conversations.

​
Permission modes

Work and Chat apply their own mode policies. Chat is read-only, and changing a permission default does not make an existing Chat run writable.

​
Settings files

Mains enables user, project, and local settings sources by default. Within these sources, local values take precedence over project values, then user values. Managed provider policy can impose additional restrictions. In a worktree, project configuration is read from that worktree’s root.

​
MCP, skills, and commands

Mains reads Claude MCP configuration from the settings files, ~/.claude.json, and the project’s .mcp.json. See MCP servers for an example. Claude discovers skills from ~/.claude/skills/ and .claude/skills/. Settings links to skills, agents, and slash-command directories. The Plugins page also supports Claude’s plugin marketplace and install scopes.

​
Subagents and output

Claude’s delegated work appears in the subagent panel. See Claude subagents and teams for the distinction between SDK subagents and interactive CLI teams. Use a structured-output schema when the task requires a defined JSON result. It is separate from creating a document or Page.