MCP Prompt and Resources

[deepakbatham572]

As per the MCP docs we can define the tools , prompts and resources and use them

Refer: MCP Prompts Doc

However , in adk we can only able to use tools with MCPToolset.
can we have something for prompts and resources as well ?

[adk-bot]

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

TLDR: You are correct. While the MCP standard includes definitions for tools, prompts, and resources, the current ADK integration primarily focuses on the 'tools' aspect via the MCPToolset. There is currently no equivalent high-level abstraction in ADK for MCP prompts and resources.

---

Hi @deepakbatham572,

Thank you for your insightful question. Your understanding of the Model Context Protocol (MCP) specification is correct; it is designed to standardize communication for tools, prompts, and resources. [2]

Your observation about the Google Agent Development Kit (ADK) is also accurate. The current implementation of the ADK-MCP integration is centered around the discovery and consumption of tools from an MCP server, which is managed through the MCPToolset class. [1] The documentation and examples show how to use MCPToolset to connect to an MCP server and load the available tools into an ADK agent. [3]

Currently, there are no equivalent high-level abstractions like MCPPromptSet or MCPResourceSet within ADK. This is a known area for future enhancement as ADK continues to evolve.

In the meantime, you can manage prompts and resources using standard Python practices within your agent's implementation:

• For Prompts: You can define your prompts as string templates or load them from configuration files within your application. These prompts can then be passed as the instruction to your agent's constructor or used dynamically within your agent's logic. [4]
• For Resources: Your application can be responsible for fetching and managing data or resources. This data can then be passed into the agent's context or supplied as arguments to your tools as needed.

We appreciate you highlighting this and it's valuable feedback for the ADK team.

---

Citations:
[1] https://google.github.io/adk-docs/tools/mcp-tools/
[2] https://google.github.io/adk-docs/mcp/
[3] https://google.github.io/adk-docs/tools/
[4] https://github.com/google/adk-python/blob/main/contributing/samples/mcp_streamablehttp_agent/agent.py