Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: While description is optional, instruction is crucial for guiding the agent on how and when to use its tools, including MCP tools. The agent can't infer the logic of tool use from tool definitions alone, so a clear instruction is necessary for effective operation.

Hello! That's a great question. Here's a breakdown of why description and instruction are important, even when using MCP tools:

• description: This is a concise summary of your agent's capabilities. [2, 9] While it's optional, it's highly recommended, especially in multi-agent setups. Other agents use the description to decide whether to delegate a task to your agent. [9]

• instruction: This is a critical parameter that guides your agent's behavior. [2] It tells the agent about its core task, its persona, any constraints it must follow, and, most importantly, how and when to use its tools. The agent needs clear instructions to understand the logic of using the tools you provide. You can even reference tools directly in the instructions and explain how to handle different return values from them. [3] The agent can't infer this logic from the tool definitions in MCP or elsewhere.

For example, even when using MCP file system tools, the description and instruction are still used to define the agent's purpose and how it should interact with those tools. [15]

In summary, while you might be able to create an agent without a description in a single-agent scenario, providing a clear and specific instruction is essential for your agent to function correctly and use its tools effectively.

I hope this helps!

[2] https://google.github.io/adk-docs/agents/llm-agents/
[3] https://google.github.io/adk-docs/tools/
[9] https://google.github.io/adk-docs/agents/multi-agents/
[15] https://google.github.io/adk-docs/tools/mcp-tools/