Skip to main content
The Claude Code agent runtime in Mains is configured through a combination of in-app settings and external configuration files. These settings control permission behavior, structured output schemas, MCP server connections, and skills discovery.

​
Settings Sources

Claude Code loads settings from multiple files, merged in priority order. Higher-priority sources override lower ones.
1

User Settings (Lowest Priority)

Path: ~/.claude/settings.json
Global settings that apply to all projects. Define your default permission mode, preferred tools, and baseline configuration here.
2

Project Settings

Path: .{your_project}/settings.json (in the project root)
Project-level settings shared with the team via version control. Use this for project-specific permission overrides, allowed tools, and MCP server definitions that all contributors should share.
3

Local Settings (Highest Priority)

Path: .{your_project}/settings.local.json (in the project root)
Personal overrides that are not committed to version control. Use this for developer-specific preferences that should not affect teammates — for example, enabling bypassPermissions on a trusted project.
Settings are merged top-down. A field defined in .{your_project}/settings.local.json takes precedence over the same field in .{your_project}/settings.json, which in turn overrides ~/.claude/settings.json.

​
MCP Servers

MCP (Model Context Protocol) servers defined in any of the settings files above are automatically loaded when the Claude Code agent starts. These servers extend the agent’s tool capabilities with custom tools, data sources, and integrations. See MCP Servers for details on configuring MCP server connections.

​
Skills

Skills are markdown instruction files that the Claude Code agent discovers and loads automatically. They provide domain-specific knowledge and behavioral instructions. Skills are loaded from two locations:
LocationScope
~/.claude/skills/Global skills available across all projects
.{your_project}/skills/Project-specific skills committed with the repository
Each skill file contains instructions that are injected into the agent’s context, giving it specialized knowledge about your codebase, conventions, or workflows.