AI Agent Toolbox

AI Agent Toolbox makes AI tool usage across models and frameworks easy. It works for parsing single use responses, or in agent and workflow loops.

Key Features

• Native support for protocols such as Anthropic MCP
• Robust parsing with streaming support
• Support for read-write and write-only tools
• Framework compatible
• Model provider-agnostic

Quick Example

from ai_agent_toolbox import Toolbox, XMLParser, XMLPromptFormatter

# Setup
toolbox = Toolbox()
parser = XMLParser(tag="use_tool")
formatter = XMLPromptFormatter(tag="use_tool")

def thinking(thoughts=""):
    print("I'm thinking:", thoughts)

# Add tools to your toolbox
toolbox.add_tool(
    name="thinking",
    fn=thinking,
    args={
        "thoughts": {
            "type": "string",
            "description": "Anything you want to think about"
        }
    },
    description="For thinking out loud"
)

system = "You are a thinking AI. You have interesting thoughts."
prompt = "Think about something interesting."

# Add our usage prompt
system += formatter.usage_prompt(toolbox)

response = llm_call(system_prompt=system, prompt=prompt)
events = parser.parse(response)

for event in events:
    toolbox.use(event)

There are many more examples in the examples folder, viewable on github - link.

Getting Started

Check out our Quick Start Guide to begin using AI Agent Toolbox.