Multiple tools error encountered in ADK multi-agent configuration and solutions

Introduction

While building an agent application using Google ADK (Agent Development Kit), I encountered the following error:

400 INVALID_ARGUMENT: Multiple tools are supported only when they are all search tools.

This article will organize the process from investigating the cause of this error to reaching a solution.

Testing environment: ADK v1.28.0, via Vertex AI, Python 3.13

Mixed Tools in a Single Agent

Initially, I was trying to create a structure that combined a custom function (Function Calling) and Google Search in a single agent.

from google.adk.agents import Agent
from google.adk.tools import google_search

agent = Agent(
    model="gemini-2.5-flash",
    name="assistant",
    instruction="...",
    tools=[
        search_knowledge_base,  # Custom Python function
        google_search,          # Gemini built-in tool
    ],
)

However, when I ran this, the following error was returned:

400 INVALID_ARGUMENT. {'error': {'code': 400, 'message': 'Multiple tools are supported only when they are all search tools.', 'status': 'INVALID_ARGUMENT'}}

Cause

Upon investigation, it appears that Gemini 2.x series models have a constraint where only one type of tool can be included in a single API request. The tools provided by ADK can be broadly categorized as follows:

All tools except custom functions are treated as Gemini built-in tools. The constraint is that you cannot include both custom functions and built-in tools in a single API request. This is a constraint of the Gemini API (model), not of the ADK.

Attempting to Workaround with sub_agents Configuration

Thinking that separating tool types might solve the issue, I changed to a sub_agents configuration where each sub-agent only has one type of tool.

search_agent = Agent(
    model="gemini-2.5-flash",
    name="search_agent",
    description="Agent that searches organization-specific knowledge base",
    instruction="...",
    tools=[search_knowledge_base],  # FunctionDeclaration only
)

web_agent = Agent(
    model="gemini-2.5-flash",
    name="web_agent",
    description="Agent that retrieves general information via Google Search",
    instruction="...",
    tools=[google_search],          # Built-in tool only
)

root_agent = Agent(
    model="gemini-2.5-flash",
    name="root_agent",
    instruction="Delegate to appropriate sub-agent based on user question",
    sub_agents=[search_agent, web_agent],
)

I thought this would work, but the same 400 INVALID_ARGUMENT error occurred with this configuration as well.

Related Issues

This problem has been reported and discussed as issues:

• google/adk-python#899
• google/adk-python#969

The issues suggest workarounds such as wrapping with AgentTool or replacing built-in tools with custom functions. However, reports indicate that results vary depending on ADK version and configuration, and these solutions didn't work in my environment.

Resolution with Gemini 3

Meanwhile, I discovered that the Gemini 3 model now supports the Tool Combination feature (preview).

https://ai.google.dev/gemini-api/docs/tool-combination?hl=ja

This allows combined use of built-in tools and custom functions within the same request.

Verification Results

Curious if this would actually work, I changed the model to gemini-3-flash-preview and tested four configurations:

With Gemini 3, it seems we no longer need to worry about tool type constraints in either single agent or sub-agent configurations.

Summary

In Gemini 2.x series, mixing built-in tools and custom functions causes a 400 INVALID_ARGUMENT error. This couldn't be avoided even by separating tools with sub_agents, which was quite a challenging constraint.

With Gemini 3's Tool Combination feature (preview), this constraint has been resolved. I've confirmed it works properly in both single agent and sub-agent configurations, so I hope this will be helpful for others facing the same error.