Agent Tools refers to a collection of external tools that AI agents can call upon to extend the capabilities of large language models. Through tool use, AI agents can perform tasks such as searching, computing, running code, manipulating files, and browsing the web — things that language models alone cannot accomplish. Agent Tools are a key infrastructure for realizing "general-purpose agents."

Core Concept

At its essence, Agent Tools provide AI agents with "hands and feet." Large language models natively only possess text generation capability; through tool calling interfaces, models can:

  • Obtain real-time information (search, database queries)
  • Execute concrete actions (run code, modify files, call APIs)
  • Perceive the external environment (web browsing, image recognition)
  • Complete complex tasks (multi-step workflow orchestration)

Main Types

Information Retrieval

  • Search engines: Web search, academic search, news queries
  • Database queries: SQL queries, vector database retrieval
  • Knowledge base access: Document retrieval, FAQ lookup

Execution & Action

  • Code execution: Python, JavaScript, and other runtime environments
  • File operations: Read/write files, directory management, format conversion
  • Shell commands: System command execution
  • API calls: Third-party service interface calls

Web & Browsing

  • Web scraping: Reading webpage content
  • Browser automation: Clicking, form filling, screenshots
  • Web interaction: Full browser session control

Multimodal

  • Image generation & recognition: Text-to-image, image-to-text, OCR
  • Audio processing: Speech recognition, speech synthesis
  • Video generation & analysis

Productivity

  • Email sending
  • Calendar management
  • Document editing
  • Spreadsheet processing

Common Protocols & Frameworks

WebMCP

WebMCP is a web-based implementation of the Model Context Protocol, allowing AI agents to interact with websites through standardized web interfaces. By providing WebMCP endpoints, websites enable AI agents to directly invoke tools and functions offered by the site — without requiring separate plugin development for each website.

Features:

  • Based on web standards, easy to deploy
  • Supports tool discovery and invocation
  • Naturally aligned with the llms.txt ecosystem
  • No additional plugins required

Function Calling / Tool Use

Native tool calling capabilities provided by major model providers:

  • OpenAI Function Calling
  • Anthropic Tool Use
  • Google Gemini Function Calling

Agent Frameworks

  • LangChain: Rich tool integrations and agent orchestration
  • AutoGPT: Autonomous task execution framework
  • CrewAI: Multi-agent collaboration framework
  • Agent Protocol: Standardized communication protocol for AI agents

MCP (Model Context Protocol)

An open protocol proposed by Anthropic, aimed at standardizing how AI models interact with external tools, enabling tools to be reused across different models and platforms.

How It Works

A typical Agent Tools invocation flow:

  1. Tool Registration: Register available tools with their names, descriptions, and parameter schemas to the AI agent
  2. Intent Recognition: The model determines which tool to call for the current task
  3. Parameter Generation: The model generates the parameters needed for the tool call
  4. Tool Execution: Actually execute the tool and obtain results
  5. Result Integration: The model integrates the tools