RK
RefundKit
GitHubGet Started

CrewAI Integration

CrewAI Integration

CrewAI is a multi-agent orchestration framework that lets you build teams of AI agents working together. This guide shows how to integrate RefundKit MCP tools into CrewAI agents for collaborative refund processing.

Prerequisites

  • Python 3.10+
  • Node.js 18+ (for the MCP server)
  • A RefundKit API key

Installation

pip install crewai crewai-tools
npm install @refundkit/sdk

Basic Setup

CrewAI supports MCP tools natively. Configure a crew with access to RefundKit:

import os
from crewai import Agent, Task, Crew, Process
from crewai_tools import MCPServerAdapter

# Configure the RefundKit MCP server
refundkit_mcp = MCPServerAdapter(
    server_params={
        "command": "npx",
        "args": ["@refundkit/sdk"],
        "env": {
            "REFUNDKIT_API_KEY": os.environ["REFUNDKIT_API_KEY"],
        },
    }
)

# Load tools from the MCP server
refundkit_tools = refundkit_mcp.tools()

# Create an agent with RefundKit tools
refund_agent = Agent(
    role="Refund Specialist",
    goal="Process customer refunds accurately and efficiently",
    backstory="""You are an experienced refund specialist. You always
    check the refund policy before processing and provide clear
    communication about the refund status and timeline.""",
    tools=refundkit_tools,
    verbose=True,
)

# Define a refund task
refund_task = Task(
    description="""Process a refund for transaction {transaction_id}.
    Amount: {amount} cents. Reason: {reason}.
    First check the policy, then process if eligible.""",
    expected_output="Refund confirmation with ID and status, or explanation of why the refund was denied.",
    agent=refund_agent,
)

# Create and run the crew
crew = Crew(
    agents=[refund_agent],
    tasks=[refund_task],
    process=Process.sequential,
    verbose=True,
)

result = crew.kickoff(
    inputs={
        "transaction_id": "ch_1N4HbSKz9cXRvFYr",
        "amount": "2500",
        "reason": "product_defective",
    }
)

print(result)

Multi-Agent Crew

CrewAI excels at multi-agent workflows. Here is a crew with specialized roles for refund processing:

# Policy analyst checks eligibility
policy_analyst = Agent(
    role="Policy Analyst",
    goal="Verify refund eligibility against company policies",
    backstory="""You are a policy analyst who checks whether refunds
    are eligible. You use the refundkit_get_policy tool to verify
    eligibility and report conditions and deadlines.""",
    tools=refundkit_tools,
)

# Refund processor handles execution
refund_processor = Agent(
    role="Refund Processor",
    goal="Execute approved refunds and track their status",
    backstory="""You process refunds that have been approved by the
    policy team. You use refundkit_process_refund to execute and
    refundkit_check_refund_status to verify completion.""",
    tools=refundkit_tools,
)

# Quality reviewer audits the results
quality_reviewer = Agent(
    role="Quality Reviewer",
    goal="Audit refund processing for accuracy and compliance",
    backstory="""You review completed refund operations to ensure
    they were processed correctly. You check amounts, reasons,
    and final statuses using refundkit_check_refund_status.""",
    tools=refundkit_tools,
)

# Define tasks for each agent
check_policy_task = Task(
    description="Check refund eligibility for transaction {transaction_id} with amount {amount} cents.",
    expected_output="Eligibility determination with policy details.",
    agent=policy_analyst,
)

process_refund_task = Task(
    description="If the policy check approved the refund, process it for transaction {transaction_id}, amount {amount} cents, reason: {reason}.",
    expected_output="Refund ID and processing status.",
    agent=refund_processor,
    context=[check_policy_task],
)

review_task = Task(
    description="Verify the processed refund was executed correctly. Check the refund status and confirm all details.",
    expected_output="Audit confirmation with refund details.",
    agent=quality_reviewer,
    context=[process_refund_task],
)

# Assemble and run the crew
crew = Crew(
    agents=[policy_analyst, refund_processor, quality_reviewer],
    tasks=[check_policy_task, process_refund_task, review_task],
    process=Process.sequential,
    verbose=True,
)

result = crew.kickoff(
    inputs={
        "transaction_id": "ch_1N4HbSKz9cXRvFYr",
        "amount": "2500",
        "reason": "product_defective",
    }
)

Restricting Tool Access

Give different agents access to different tools for security:

# Only give the policy analyst the policy tool
policy_tools = [t for t in refundkit_tools if "policy" in t.name]

# Only give the processor the process and cancel tools
processor_tools = [t for t in refundkit_tools if "process" in t.name or "cancel" in t.name]

Next Steps