Tool
What is a tool in AI?
In the context of AI agents, tools are capabilities that extend an agent's abilities beyond text generation by allowing the agent to request specific actions—like executing code, calling APIs, or accessing databases—and receiving responses that can inform subsequent actions.
Tools are defined by the developer and made available to the agent, which can then decide when and how to use them based on the user's input. This ability to invoke external capabilities distinguishes AI agents from simpler chatbots or workflows.
How do tools work with AI agents?
When building an AI agent, developers implement specific tools and make them available to the agent. The agent can then autonomously decide which tools to use and when, based on the context of the conversation.
For example, a customer interview coaching agent might have access to these tools:
- Student Course Lookup: Retrieve a student's course history
- Student Transcript Lookup: Retrieve previous interview transcripts and coaching responses
- Update Student Transcript History: Save the latest feedback to the student's record
The agent would autonomously decide whether to look up the student's history before providing feedback, then update their record after giving coaching tips.
What is the Model Context Protocol?
The Model Context Protocol (MCP) is an open standard defined by Anthropic for connecting agents to tools. MCP makes it easy for agents to discover what tools are available, how to authenticate to use those tools, and how to call them.
With MCP, tools can be developed and augmented separately from the agent itself. Each time the agent connects to an MCP server, it retrieves the current list of available tools, enabling flexible tool development without modifying the core agent.
Learn more:
Related terms: