Built-in Tools Guide

LibrAgent comes with an out-of-the-box set of native Built-in Tools, enabling agents to inspect files, edit code, browse the web, capture screenshots, and execute terminal commands immediately without requiring external tool setups.

When you send a request, the agent autonomously selects and invokes the most appropriate tools to complete your task.

---

🛡️ Safety & Execution Approval Policies

Tool execution remains strictly under user supervision:

Tool Category	Safety Policy	Key Operations
Read-Only Tools	Automatic execution	Reading files, viewing web pages, searching session history
System Modification Tools	Approval Dialog (Normal mode)	Running shell commands (runShell), creating or modifying files
Desktop Control Tools	Explicit Approval Required	Taking desktop screenshots (captureScreen), simulating mouse/keyboard (computerControl)

TIP

For unattended background runs (e.g., Scheduled Tasks), you can switch the execution mode to YOLO or Unsafe to auto-approve standard tool executions without blocking on confirmation popups.

---

🛠️ Core Tool Capabilities

1. Workspace & Files (workspace)

Agents inspect and edit files directly within your configured project directory.

• Reading & Searching: Understands codebase structure and text documents.
• Precise Line Edits: Makes surgical replacements without overwriting entire files.
• Terminal Execution: Runs builds, test suites, or package managers and captures output.

2. Web Browsing (browser)

Interacts with the web through an isolated browser process.

• Web Navigation: Reads online documentation, technical articles, and live feeds.
• Visual Verification: Captures screenshots of rendered web pages.

3. Media & Desktop Interaction (media, desktop)

Handles visual analysis and operating system interactions.

• Screen Capture: Inspects error windows or UI layouts visually.
• Mouse & Keyboard Control: Simulates clicks, typing, and shortcuts for GUI workflows. (Always requires manual approval)

4. Planning & Scheduling (planning, scheduled_task)

Organizes complex goals into structured sub-tasks.

• Multi-step Goals: Breaks down large instructions into a structured Todo list and reports progress.
• Scheduled Automations: Registers recurring background tasks according to Cron expressions.

---

⚙️ Optimizing Tools per Assistant

Enabling every tool in all sessions increases the context tokens the LLM must consume to read tool definitions.

1. Navigate to Assistants in the sidebar.
2. Select Edit → Tools on any assistant profile.
3. Keep only relevant tools active (e.g., enable workspace for coding assistants, browser for research assistants).

---

📖 Technical Reference: Tool Identifiers ({server}__{tool})

For power users and prompt engineering, below is the primary built-in tools inventory:

Server	Tool Identifier	Description	Default Status
workspace	workspace__readFile	Slice and read file contents	Core (Default)
	workspace__writeFile	Create or overwrite files	Core (Default)
	workspace__strReplace	Exact string replacement in file	Core (Default)
	workspace__listDirectory	List files and directories	Core (Default)
	workspace__runShell	Execute shell command (async supported)	Core (Default)
browser	browser__navigateToUrl	Navigate to URL and wait for load	Optional
	browser__takeScreenshot	Capture viewport or full page screenshot	Optional
	browser__getPageContent	Extract webpage text & DOM structure	Optional
	browser__clickElement	Click DOM element	Optional
desktop	desktop__computerControl	Mouse clicks, movement, and keyboard typing	Optional (Approval required)
media	media__captureScreen	Capture desktop display or area	Optional (Approval required)
	media__seeContent	Analyze visual image/video content	Optional
	media__listenContent	Analyze audio/video soundtrack	Optional
planning	planning__createGoal	Create multi-step goals & todos	Optional
	planning__updateGoal	Update goal & step progress	Optional
scheduled_task	scheduled_task__createScheduledTask	Register recurring background Cron task	Core (Default)
	scheduled_task__listScheduledTasks	List registered scheduled tasks	Core (Default)

On successful context compaction, planning todos are cleared and the goal is kept. Rebuild the checklist from the compact handoff with planning__addTodo if steps remain. A successful ui__reportResult clears goal, todos, and scratchpad so the next request starts clean.

---

🧩 Need Additional Tools?

If you require integration with third-party services (GitHub, Slack, remote databases):

• Visit the Extensions Guide to add verified MCP presets with one click or register custom MCP servers.