Conditionally enable tools based on tool call history

[dannyhorcajo]

Please read this first

• Have you read the docs? Agents SDK docs
• Have you searched for related issues? Others may have had similar requests

Question

I would like to dynamically enable certain tools / MCP server tools depending on whether other tools have been previously invoked. For example, query_db() should only be enabled if the agent has invoked the get_db_schema() tool previously during the conversation.

Is such behavior possible with this SDK? I have been thinking of using the run context for this, but it would get reset at each turn. I have been reading the documentation from A to Z, but I couldn't find anything of the kind.

Thank in advance!

[seratch]

Although agents as tools / handoffs of agents provide is_enabled function feature (see https://openai.github.io/openai-agents-python/tools/#conditional-tool-enabling and https://openai.github.io/openai-agents-python/handoffs/), function tools and MCP server tools do not have the same (and we don't have immediate plans to add the same to these).

General suggestions are:

1. Enhance your instructions (aka system prompt) to let agents better understand your goal and requirements
2. Run a planner agent first, and then manually call required agents separately: https://github.com/openai/openai-agents-python/blob/main/examples/financial_research_agent/manager.py

The second one may not look elegant, but it would be the most predictable design for scenarios like yours. I hope this helps!

[dannyhorcajo]

Got it! I already had those approaches in mind, but I wanted to confirm that the feature did indeed not exist.

Thanks for the quick answer!

[ihower]

@dannyhorcajo Actually, I found that regular @function_tool do support the is_enabled parameter — so it’s possible to dynamically enable or disable tools based on the RunContextWrapper data.

Here’s a simple example: the bar tool will only be available after the foo tool has been invoked.

from dataclasses import dataclass
from agents import Agent, function_tool, Runner,RunContextWrapper, AgentBase

def is_bar_enabled(ctx: RunContextWrapper, agent: AgentBase) -> bool:
    """Check if bar tool should be enabled based on whether foo was called."""
    return ctx.context.executed_tools.get("foo", False)

@function_tool
async def foo(ctx: RunContextWrapper) -> str:
    ctx.context.executed_tools["foo"] = True
    return "41"

@function_tool(is_enabled=is_bar_enabled)
async def bar() -> str:
    return "64"

@dataclass
class RunContext:
    executed_tools: dict[str, bool]

agent = Agent[RunContext](
    name="Assistant",
    instructions="You are a helpful assistant. 用中文回答.",
    model="gpt-4.1-mini",
    tools=[foo, bar]
)

async def main():
    """Main function demonstrating conditional tool availability."""
    run_context = RunContext(executed_tools={})

    print("-------------------------------- Round 1 --------------------------------")

    messages = [{"role": "user", "content": "Call the foo tool first. After getting the result, you can call the bar tool, and finally tell me the result."}]

    result = await Runner.run(agent, input=messages, context=run_context)

    print(result.final_output)
    
    print("-------------------------------- Round 2 --------------------------------")
    
    messages = result.to_input_list()

    messages.append({"role": "user", "content": "Call the bar tool again and tell me the result"})
    result = await Runner.run(agent, input=messages, context=run_context)
    
    print(result.final_output)

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())

# Result:
# -------------------------------- Round 1 --------------------------------
# 1st API call: Calls foo tool. Available tools: [foo]
# 2nd API call: Calls bar tool. Available tools: [foo, bar]
# 3rd API call: Returns the final text answer

# -------------------------------- Round 2 --------------------------------
# 1st API call: Calls bar tool. Available tools: [foo, bar]
# 2nd API call: Returns the final text answer