CrewAI Tax Compliance

3-step setup

Define the tool

CrewAI uses BaseTool with Pydantic schemas for typed input validation. This keeps the agent's arguments structured and predictable.

from crewai.tools import BaseTool
from pydantic import BaseModel, Field
from agenttax import AgentTaxClient

client = AgentTaxClient(api_key="atx_live_...")

class TaxInput(BaseModel):
    amount: float = Field(description="Transaction amount in USD")
    buyer_state: str = Field(description="Two-letter US state code")
    transaction_type: str = Field(description="saas, compute, consulting, research, content")
    counterparty_id: str = Field(description="Unique ID of the buyer")
    is_b2b: bool = Field(default=False, description="Is this a B2B transaction?")

class SalesTaxTool(BaseTool):
    name: str = "calculate_sales_tax"
    description: str = (
        "Calculate US sales tax for a transaction. Returns the tax amount, "
        "combined rate, taxability, and confidence score. Use this before "
        "generating any invoice or payment request."
    )
    args_schema: type[BaseModel] = TaxInput

    def _run(self, amount, buyer_state, transaction_type, counterparty_id, is_b2b=False):
        result = client.calculate(
            role="seller",
            amount=amount,
            buyer_state=buyer_state,
            transaction_type=transaction_type,
            counterparty_id=counterparty_id,
            is_b2b=is_b2b,
        )
        return (
            f"Tax result for ${amount} {transaction_type} in {buyer_state}: "
            f"taxable={result['taxable']}, total_tax=${result['total_tax']:.2f}, "
            f"rate={result['combined_rate']*100:.2f}%, "
            f"confidence={result['confidence']['level']}"
        )

tax_tool = SalesTaxTool()

Single-agent crew

A billing agent that calculates tax before invoicing.

from crewai import Agent, Task, Crew

billing_agent = Agent(
    role="Billing Specialist",
    goal="Calculate the correct tax and generate accurate invoices for agent services",
    backstory="Expert in AI agent commerce tax obligations across all 50 US states",
    tools=[tax_tool],
    verbose=True,
)

tax_task = Task(
    description=(
        "Calculate the sales tax for this transaction: "
        "$2,500 SaaS service sold to agent-client-99 in New York. "
        "B2B transaction. Determine tax owed and generate the invoice total."
    ),
    expected_output="Invoice total including itemized tax breakdown",
    agent=billing_agent,
)

crew = Crew(agents=[billing_agent], tasks=[tax_task])
result = crew.kickoff()
# Agent calls calculate_sales_tax(2500, "NY", "saas", "agent-client-99", True)
# NY SaaS B2B: taxable, 4% state rate → $100 tax → total $2,600

Multi-agent: tax validator crew

Separate validator and sales agents — each with the tax tool. Good for high-value transactions requiring independent verification.

# Multi-agent crew: one agent sells, one validates tax compliance

tax_validator = Agent(
    role="Tax Compliance Validator",
    goal="Verify that every outbound invoice correctly reflects tax obligations",
    tools=[tax_tool],
)

sales_agent = Agent(
    role="Sales Agent",
    goal="Sell compute services and generate invoices",
    tools=[tax_tool],
)

validation_task = Task(
    description="Validate the tax calculation on invoice #4821 ($5,000 compute, TX, B2B)",
    agent=tax_validator,
)

# TX compute B2B: 80% taxable rule → $5,000 × 80% × 6.25% = $250 tax

A crew that transacts on a customer’s behalf also has to answer who the seller is, which the tool cannot decide for you. The operator, facilitator and merchant-of-record decision tree runs that question in the order it has to be run.

What your crew gets

Other integrations