<aside> 🦥

Sources

This is mainly from: https://cdn.openai.com/business-guides-and-resources/a-practical-guide-to-building-agents.pdf

</aside>

Core Categories

<aside> 🦥

• Data Tools: Enable agents to retrieve context and information needed for workflow execution
  • Query transaction databases or CRM systems
  • Read PDF documents
  • Search the web for information
  • Retrieve knowledge base articles </aside>

<aside> 🦥

• Action Tools: Allow agents to interact with systems to make changes or take actions
  • Send emails and text messages
  • Update CRM records
  • Hand off customer service tickets to humans
  • Initiate payments or refunds
  • Commit code changes </aside>

<aside> 🦥

• Orchestration Tools: Other agents that can be used as tools by a manager agent
  • Specialized agents (e.g., Refund agent, Research agent, Writing agent)
  • Allow for delegation of specific tasks to purpose-built agents </aside>

Tool Implementation Considerations

• Standardize tool definitions for flexible, many-to-many relationships between tools and agents
• Create well-documented, thoroughly tested, and reusable tools
• Improve discoverability and prevent redundant definitions
• Rate tools by risk level (low, medium, high) based on their potential impact
• Consider factors like read-only vs. write access, reversibility, and financial impact

Tool Safeguards

• Implement guardrails to prevent misuse of high-risk tools
• Pause for additional checks before executing sensitive functions
• Trigger human approval for tools that could have significant consequences
• Consider risk factors when determining which tools to make available to which agents

Legacy System Integration

• For systems without APIs, use computer-use models to interact directly with UIs
• Allow agents to interact with web and application interfaces just as humans would

🔌 The Modern Way to Connect Tools: MCP