Tools¶

Tools are what make Claude Code agentic. Without tools, Claude can only respond with text. With tools, Claude can act: read code, edit files, run commands, search the web, and interact with external services.

Each tool use returns information that feeds back into the agentic loop, informing Claude's next decision.

The agentic loop¶

When you give Claude a task, it works through three phases: gather context, take action, and verify results. Claude uses tools throughout all phases, chaining dozens of actions together and course-correcting along the way.

You can interrupt at any point to steer Claude in a different direction.

Built-in tools¶

File operations¶

Tool	Description	Permission
Read	Read file contents	No
Edit	Targeted edits to specific files	Yes
Write	Create or overwrite files	Yes
NotebookEdit	Edit Jupyter notebook cells	Yes

Search¶

Tool	Description	Permission
Glob	Find files by pattern matching (e.g., `*/.yaml`)	No
Grep	Search file contents with regex	No

Execution¶

Tool	Description	Permission
Bash	Execute shell commands in your environment	Yes

Web¶

Tool	Description	Permission
WebSearch	Search the web for information	No
WebFetch	Fetch and process content from a URL	No

Orchestration¶

Tool	Description	Permission
Task	Spawn subagents for delegated work	No
TaskOutput	Retrieve output from background tasks	No
AskUserQuestion	Ask the user for clarification	No
ExitPlanMode	Exit plan mode and start implementing	Yes

Extending tools¶

The built-in tools are the foundation. Claude Code can be extended with additional tools through:

MCP servers: Connect external services (databases, APIs, custom tools) via the Model Context Protocol. See MCP servers
Skills: Domain-specific workflows that orchestrate tool usage. See Agent skills
Subagents: Delegated workers with their own tool access and context. See Subagents
Hooks: Shell commands that run automatically on tool events. See Hooks

Permissions¶

Tools that modify state (Edit, Write, Bash) require permission. Claude asks before using them unless you pre-approve in settings.json.

Permission modes¶

Cycle through modes with Shift+Tab:

Default: Claude asks before file edits and shell commands
Auto-accept edits: Claude edits files without asking, still asks for commands
Plan mode: Read-only tools only, creates a plan for approval

Permission rules in settings¶

{
  "permissions": {
    "allow": [
      "Bash(kubectl get:*)",
      "Bash(npm run *)",
      "Read"
    ],
    "deny": [
      "Bash(kubectl delete:*)",
      "Read(./.env)"
    ]
  }
}

Rules support glob patterns for command arguments. Deny rules take precedence over allow rules.

How Claude chooses tools¶

Claude selects tools based on the prompt and what it learns at each step. For example, when asked to "fix the failing tests", Claude might:

Bash - Run the test suite to see failures
Read - Read the error output and relevant source files
Edit - Fix the code
Bash - Run tests again to verify

Each tool use returns information that informs the next step. This is the agentic loop in action.

Links¶

How Claude Code works

https://code.claude.com/docs/en/how-claude-code-works

Settings and permissions

https://code.claude.com/docs/en/settings