File & Code Tools
FileReadTool
Reads file contents with support for: plain text files (with line numbers), PDF documents (page-by-page extraction), Jupyter notebooks (.ipynb with cell outputs), images (PNG, JPG - visual display for multimodal LLM), binary detection and rejection. Supports line limiting (offset + limit) for large files, and compresses images to reduce token usage.
Parameters:
| Name | Type | Required | Description |
|---|---|---|---|
file_path | string | Yes | Absolute path to the file |
offset | number | No | Line number to start reading from |
limit | number | No | Number of lines to read (default: 2000) |
pages | string | No | PDF page range (e.g., "1-5") |
FileEditTool
Performs exact string replacements in files. Uses old_string/new_string matching (must be unique in file). Supports replace_all for global renaming. Shows diff preview before applying. Requires file to be read first (via FileReadTool) to prevent blind edits.
Parameters:
| Name | Type | Required | Description |
|---|---|---|---|
file_path | string | Yes | Absolute path to the file to modify |
old_string | string | Yes | The text to replace (must be unique) |
new_string | string | Yes | The replacement text |
replace_all | boolean | No | Replace all occurrences (default: false) |
FileWriteTool
Creates or overwrites files atomically. Checks permissions before writing. Validates parent directory exists. Will not create files without reading the existing file first (if it exists).
GlobTool
Fast file pattern matching using glob syntax. Returns matching file paths sorted by modification time. Works with any codebase size.
GrepTool
Powered by ripgrep. Full regex support. Output modes: content (with context lines), files_with_matches, count. Supports file type and glob filtering.
LSPTool
Language Server Protocol integration providing IDE features: go-to-definition, hover info, symbol search, completion, diagnostics.
Go to the Tool Simulator in the Playground. Select FileReadTool and enter a file path. Watch how it validates the path and returns line-numbered content.
Open Tool Simulator →Execution Tools
BashTool
The most complex tool in Claude Code. Executes shell commands with real-time streaming output, environment variable control, timeout management, background execution mode, and a massive 98KB security validation module.
Security Sub-Modules:
bashSecurity.ts (98KB) - Main command validation enginepathValidation.ts (43KB) - Path access controlreadOnlyValidation.ts (68KB) - Read-only mode enforcementsedValidation.ts - Sed script safety analysisdestructiveCommandWarning.ts - Warns before dangerous opsshouldUseSandbox.ts - Sandbox decision logiccommandSemantics.ts - Command intent analysisPowerShellTool
Windows-specific shell execution. Loaded conditionally only on Windows systems.
NotebookEditTool
Jupyter notebook cell operations: create, edit, delete, move cells, change cell type. Preserves notebook metadata and outputs.
Go to the Security Command Tester. Type 'rm -rf /' and see how bashSecurity.ts catches it. Then try 'ls -la' and see it pass. Can you find a command that looks safe but gets blocked?
Open Security Tester →Agent & Team Tools
AgentTool
Spawns independent sub-agents with their own conversation context. Supports 5 built-in agent types with foreground/background execution and worktree isolation.
Built-in Agent Types:
| Agent | Purpose | Tools |
|---|---|---|
general-purpose | Complex multi-step tasks | All tools |
Explore | Fast codebase search | Read-only (no Edit/Write) |
Plan | Architecture design | Read-only + planning |
claude-code-guide | Help & documentation | Glob, Grep, Read, WebFetch |
verification | Test running | Build + test tools |
TeamCreateTool / TeamDeleteTool / SendMessageTool
Internal-only tools for multi-agent team orchestration. Not available in normal mode.
Think about how you'd parallelize searching a large codebase. You'd spawn 3 Explore agents for different directories - each runs independently. This is exactly what AgentTool does.
Open Tool Simulator →Web & Search Tools
WebSearchTool
Web search via Claude's native web_search tool.
WebFetchTool
HTTP fetch with automatic HTML-to-Markdown conversion.
ToolSearchTool
Discovers available tools dynamically via deferred loading.
ListMcpResourcesTool / ReadMcpResourceTool
MCP resource discovery and reading from connected servers.
Task & Schedule Tools
TaskCreateTool
Creates structured task entries with status tracking.
TaskUpdateTool
Updates task status, dependencies, and metadata.
TaskGetTool / TaskListTool / TaskStopTool / TaskOutputTool
Complete task lifecycle management.
CronCreateTool / CronDeleteTool / CronListTool
Cron job management for scheduled recurring tasks.
RemoteTriggerTool
Schedules remote async agent execution on Anthropic's infrastructure.
In the Playground Token Calculator, set 100 messages and check the budget. Now imagine each TaskCreate adds ~50 tokens overhead. How does that affect your budget?
Open Token Calculator →Control & Mode Tools
EnterPlanModeTool / ExitPlanModeV2Tool
Plan mode enables extended thinking with 30-minute budget for complex architectural decisions.
EnterWorktreeTool / ExitWorktreeTool
Git worktree management for isolated experimentation.
AskUserQuestionTool
Prompts user for input when AI needs clarification.
SkillTool
Executes MCP skills and custom /slash commands.
Go to the Permission Simulator. Select BashTool:git push --force in Bypass mode. See how it gets auto-approved with NO safety check. Now switch to Default mode - see the difference?
Open Permission Simulator →