artificer-agents 0.1.0a4

A library for creating agents

artificer-agents

A lightweight Python library for running deterministic agent loops with MCP tools.

Installation

pip install artificer-agents

# With OpenAI support
pip install artificer-agents[openai]

Requires Python 3.13+.

Usage

Define agents as classes:

import asyncio
from pydantic import BaseModel
from artificer.agents import Agent
from artificer.mcp import MCPClient
from artificer.models import OpenAIModel


class Input(BaseModel):
    query: str


class Output(BaseModel):
    result: str


class MyAgent(Agent):
    system_prompt = "You are a helpful assistant."
    model = OpenAIModel(model="gpt-4o-mini")
    mcp_client = MCPClient(["path/to/mcp-server.py"])  # optional
    input_schema = Input
    output_schema = Output


async def main():
    agent = MyAgent()

    async with agent.mcp_client:
        result = await agent.run(Input(query="Hello"), verbose=True)
        print(result.result)

asyncio.run(main())

Simple String Agents

For simple str -> str agents, no schemas needed:

class SimpleAgent(Agent):
    system_prompt = "Answer questions concisely."
    model = OpenAIModel(model="gpt-4o-mini")

result = await SimpleAgent().run("What is Python?")
print(result.output)  # StringOutput with .output field

Subagents

Agents can spawn other agents:

class ResearcherAgent(Agent):
    system_prompt = "Research the topic and return findings."
    model = model
    mcp_client = mcp_client
    input_schema = ResearchInput
    output_schema = ResearchOutput


class OrchestratorAgent(Agent):
    system_prompt = "Use the researcher subagent to gather info."
    model = model
    subagents = [ResearcherAgent()]
    input_schema = Input
    output_schema = Output

Development

uv sync                    # Install dependencies
./scripts/check.sh         # Run all checks (lint, format, typecheck, tests)
./scripts/test.sh          # Run tests only
./scripts/format.sh        # Format code

License

MIT