Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.getovra.com/llms.txt

Use this file to discover all available pages before exploring further.

CrewAI orchestrates specialized agents as a team — purchaser, auditor, reporter. Ovra’s MCP server connects natively; each role gets the same tool surface but with different instructions and goals.
Sandbox-only today. Use a sk_sandbox_* or sk_test_* key.

Install

pip install crewai mcp

Setup

from crewai import Agent, Task, Crew
from crewai.mcp import MCPServerHTTP

ovra_server = MCPServerHTTP(
    url="https://api.getovra.com/api/mcp",
    headers={"Authorization": "Bearer sk_sandbox_..."},
    streamable=True,
)

purchasing = Agent(
    role="Purchasing Agent",
    goal="Find and buy products under budget",
    backstory="Expert online shopper with payment authority",
    mcps=[ovra_server],
)

auditor = Agent(
    role="Compliance Auditor",
    goal="Verify every purchase complies with policy and is receipt-attached",
    backstory="Financial-controls specialist",
    mcps=[ovra_server],
)

Multi-agent workflow

purchase_task = Task(
    description="Buy a USB-C hub under EUR 30 on amazon.de",
    expected_output="Order confirmation with transaction ID",
    agent=purchasing,
)

audit_task = Task(
    description=(
        "Verify the purchase complies with policy and attach the receipt. "
        "Surface any policy advisories."
    ),
    expected_output="Compliance report with intent ID, transaction ID, and policy decision",
    agent=auditor,
)

crew = Crew(
    agents=[purchasing, auditor],
    tasks=[purchase_task, audit_task],
    verbose=True,
)

result = crew.kickoff()

Tool usage by role

Purchasing:
  ovra_intent { declare }
  ovra_pay    { checkout }   <- MPP-aware merchant
  ovra_intent { verify }

Auditor:
  ovra_transaction { get }
  ovra_policy      { get }
  ovra_receipt     { upload }   <- attaches PDF evidence
  ovra_outcome     { report }

Why CrewAI

  • Role-based agents — separate purchasing, auditing, reporting concerns
  • Policy enforcement — Ovra validates every charge regardless of which agent triggered it
  • Receipt chain — auditor attaches receipts; chain is audit-logged
ToolPurpose
ovra_payFull flow in one call
ovra_intentDeclare and verify
ovra_credentialFine-grained lifecycle
ovra_transactionAudit history
ovra_receiptAttach evidence
ovra_policyRead before declaring
ovra_outcomeReport results

Next

MCP overview

Architecture and the full 19-tool list.

OpenAI Agents

Hosted MCP integration.

LangGraph

Stateful graph workflows.