ContextQMD
Back to UI-TARS-desktop

Tool Development

multimodal/websites/tarko/docs/en/guide/tool/basic.mdx

0.3.05.4 KB
Original Source

Tool Development

Learn how to develop custom tools to extend Agent capabilities.

Overview

Tools are the building blocks that enable Agents to interact with external systems. @agent-tars/core uses the MCP (Model Context Protocol) framework for tool integration, providing a standardized way to create and register tools.

Tool Architecture

Tools in @agent-tars/core are based on MCP Tool interface from @tarko/mcp-agent:

typescript
import { Tool } from '@tarko/mcp-agent';

// Tools are created using the Tool class constructor
const tool = new Tool({
  id: 'tool_name',
  description: 'Tool description',
  parameters: JSONSchema7,
  function: async (args) => { /* implementation */ }
});

Creating Your First Tool

Basic Tool Structure

typescript
import { Tool, JSONSchema7 } from '@tarko/mcp-agent';

export function createCalculatorTool(): Tool {
  return new Tool({
    id: 'calculator',
    description: 'Perform basic mathematical calculations',
    parameters: {
      type: 'object',
      properties: {
        operation: {
          type: 'string',
          enum: ['add', 'subtract', 'multiply', 'divide'],
          description: 'Mathematical operation to perform'
        },
        a: {
          type: 'number',
          description: 'First number'
        },
        b: {
          type: 'number',
          description: 'Second number'
        }
      },
      required: ['operation', 'a', 'b']
    } as JSONSchema7,
    function: async (params: {
      operation: string;
      a: number;
      b: number;
    }) => {
      const { operation, a, b } = params;
      
      let result: number;
      
      switch (operation) {
        case 'add':
          result = a + b;
          break;
        case 'subtract':
          result = a - b;
          break;
        case 'multiply':
          result = a * b;
          break;
        case 'divide':
          if (b === 0) {
            throw new Error('Division by zero is not allowed');
          }
          result = a / b;
          break;
        default:
          throw new Error(`Unknown operation: ${operation}`);
      }
      
      return [{
        type: 'text',
        text: `${a} ${operation} ${b} = ${result}`
      }];
    }
  });
}

Registering Tools

Tools are registered through MCP servers or directly in AgentTARS:

typescript
import { AgentTARS } from '@agent-tars/core';
import { createCalculatorTool } from './calculator-tool';

// Register tools through custom MCP server
const agent = new AgentTARS({
  mcpServers: {
    calculator: {
      // Custom MCP server with calculator tool
    }
  }
});

// Or register during initialization
agent.registerTool(createCalculatorTool());

Built-in Tool Categories

@agent-tars/core includes built-in tools across 4 MCP server categories:

  • browser: Web automation via @agent-infra/mcp-server-browser
  • filesystem: File operations via @agent-infra/mcp-server-filesystem
  • commands: System commands via @agent-infra/mcp-server-commands
  • search: Web search via @agent-infra/search

These are managed through BrowserToolsManager, FilesystemToolsManager, and SearchToolProvider classes.

Tool Implementation Patterns

Async Operations

MCP tools naturally support async operations:

typescript
export function createApiTool(): Tool {
  return new Tool({
    id: 'api_request',
    description: 'Make HTTP API requests',
    parameters: {
      type: 'object',
      properties: {
        url: { type: 'string', description: 'Request URL' },
        method: { type: 'string', enum: ['GET', 'POST', 'PUT', 'DELETE'] }
      },
      required: ['url', 'method']
    } as JSONSchema7,
    function: async (params: { url: string; method: string }) => {
      try {
        const response = await fetch(params.url, {
          method: params.method
        });
        
        const data = await response.text();
        
        return [{
          type: 'text',
          text: `Status: ${response.status}\n${data}`
        }];
      } catch (error) {
        throw new Error(`API request failed: ${error.message}`);
      }
    }
  });
}

Best Practices

Error Handling

Use proper error handling with meaningful messages:

typescript
function: async (params: any) => {
  try {
    // Tool logic here
    return [{ type: 'text', text: 'Success result' }];
  } catch (error) {
    throw new Error(`Tool execution failed: ${error.message}`);
  }
}

Parameter Validation

Define comprehensive JSON Schema for parameters:

typescript
parameters: {
  type: 'object',
  properties: {
    email: {
      type: 'string',
      format: 'email',
      description: 'Valid email address'
    },
    age: {
      type: 'number',
      minimum: 0,
      maximum: 150,
      description: 'Age in years'
    }
  },
  required: ['email'],
  additionalProperties: false
} as JSONSchema7

Testing Tools

Test your tools using standard testing frameworks:

typescript
import { createCalculatorTool } from './calculator-tool';

describe('Calculator Tool', () => {
  it('should perform calculations', async () => {
    const tool = createCalculatorTool();
    
    const result = await tool.function({
      operation: 'add',
      a: 5,
      b: 3
    });
    
    expect(result[0].text).toContain('8');
  });
});

Next Steps