GuideMulti-Agent FrameworksPython SDK

AI Agent Framework Integrations

Equip autonomous agents with zero-retention 2FA, OTP, Magic-Link, and TOTP authenticator vault tools. Native integrations for Google ADK, Microsoft Agent Framework, PydanticAI, OpenAI Agents SDK, Claude Agent SDK, Strands Agents, Agno, Smolagents, LangGraph, and CrewAI.

The Autonomous 2FA Bottleneck

Autonomous agents navigating login portals, SaaS onboarding, or high-security actions inevitably hit 2FA barriers. Without an inbox API, agents hang or trigger catastrophic polling loops. Sentry Relay eliminates this bottleneck by pairing unique, ephemeral Identities with a sub-200ms Redis long-polling wait primitive and an RFC 6238 TOTP vault.

Sub-200ms /wait

Long-polls via Redis pub/sub. Immediately unblocks the agent the moment the SMS or email OTP arrives.

Expiring Window Protection

Automatically pauses if ≤4 seconds remain in a 30s TOTP period, preventing stale code submission failures.

Zero-PII & Ephemeral

Codes are extracted in memory and purged automatically. No raw emails or SMS text are persisted to disk.

Installation & Optional Extras

Install the core Python SDK or include extras tailored to your target agent framework:

# Core SDK
pip install sentryrelay-sdk

# Framework-specific extras:
pip install "sentryrelay-sdk[google-adk]"        # Google ADK
pip install "sentryrelay-sdk[microsoft-agents]"   # Microsoft Agent Framework
pip install "sentryrelay-sdk[pydantic-ai]"        # PydanticAI
pip install "sentryrelay-sdk[openai-agents]"      # OpenAI Agents SDK
pip install "sentryrelay-sdk[claude-agent]"       # Claude Agent SDK / Anthropic
pip install "sentryrelay-sdk[strands]"            # Strands Agents
pip install "sentryrelay-sdk[agno]"               # Agno
pip install "sentryrelay-sdk[smolagents]"         # Smolagents
pip install "sentryrelay-sdk[langchain]"          # LangGraph / LangChain
pip install "sentryrelay-sdk[crewai]"             # CrewAI

Framework Recipes

Copy-pasteable integrations for all major LLM agent frameworks.

Google ADK (Agent Development Kit)

Provides ready-to-use tool callables for Google ADK LLM agents using Gemini and Vertex AI models:

python

from google.adk.agents.llm_agent import Agent
from sentryrelay import SentryRelay
from sentryrelay.integrations.google_adk import create_google_adk_tools

relay = SentryRelay()
tools = create_google_adk_tools(relay)

agent = Agent(
    model="gemini-2.5-flash",
    name="checkout_verifier",
    instruction="Solve 2FA challenges by waiting on inbound codes or retrieving TOTPs.",
    tools=tools,
)

Microsoft Agent Framework

Compatible with Microsoft Agent Framework (agent-framework) and Azure AI Projects. Automatically wraps functions with AIFunction:

python

from agent_framework import ChatAgent
from sentryrelay import SentryRelay
from sentryrelay.integrations.microsoft_agent import create_microsoft_agent_tools

relay = SentryRelay()
tools = create_microsoft_agent_tools(relay)

agent = ChatAgent(
    name="SecurityAgent",
    tools=tools,
    instructions="Handle 2FA and OTP authentication challenges autonomously.",
)

PydanticAI

Type-safe agent tools with structured schemas. Seamlessly register tools directly onto your agent:

python

from pydantic_ai import Agent
from sentryrelay import SentryRelay
from sentryrelay.integrations.pydantic_ai import register_sentryrelay_tools

relay = SentryRelay()
agent = Agent('openai:gpt-4o', system_prompt="Handle automated authentication.")

# Registers wait_for_verification_code, get_totp_code, and list_identities
register_sentryrelay_tools(agent, relay)

result = agent.run_sync("Log into Okta using identity dev-agent and provide the OTP.")
print(result.data)

OpenAI Agents SDK

For the official OpenAI Agents SDK (openai-agents / agents):

python

from agents import Agent, Runner
from sentryrelay import SentryRelay
from sentryrelay.integrations.openai_agents import create_openai_agents_tools

relay = SentryRelay()
tools = create_openai_agents_tools(relay)

agent = Agent(
    name="LoginAssistant",
    instructions="You automate account sign-ins. Use Sentry Relay tools to solve 2FA challenges.",
    tools=tools,
)

result = Runner.run_sync(agent, "Sign in using checkout-bot and retrieve the incoming SMS code.")
print(result.final_output)

Claude Agent SDK & Anthropic Tool Use

Provides complete Anthropic tool schema definitions (tools.definitions) and a unified dispatcher (tools.execute(name, args)):

python

import anthropic
from sentryrelay import SentryRelay
from sentryrelay.integrations.claude_agent import create_claude_agent_tools

relay = SentryRelay()
tools = create_claude_agent_tools(relay)
client = anthropic.Anthropic()

# Supply tools.definitions directly to messages.create
response = client.messages.create(
    model="claude-3-7-sonnet-20250219",
    max_tokens=1024,
    tools=tools.definitions,
    messages=[{"role": "user", "content": "Wait for OTP code on identity 'checkout-bot'"}],
)

# Dispatch execution
for block in response.content:
    if block.type == "tool_use":
        result = tools.execute(block.name, block.input)
        print("Tool Output:", result)

Strands Agents

Decorated with @tool from strands:

python

from strands import Agent
from sentryrelay import SentryRelay
from sentryrelay.integrations.strands import create_strands_tools

relay = SentryRelay()
tools = create_strands_tools(relay)

agent = Agent(
    model="claude-3-7-sonnet-20250219",
    name="browser_auth_agent",
    system_prompt="Resolve OTP and TOTP verification challenges using Sentry Relay.",
    tools=tools,
)

Agno (formerly Phidata)

A native Toolkit subclass for Agno agents:

python

from agno.agent import Agent
from agno.models.openai import OpenAIChat
from sentryrelay import SentryRelay
from sentryrelay.integrations.agno import SentryRelayTools

relay = SentryRelay()

agent = Agent(
    model=OpenAIChat(id="gpt-4o"),
    tools=[SentryRelayTools(relay)],
    instructions="Solve 2FA challenges by waiting on inbound codes or generating live TOTPs.",
)

agent.print_response("Wait for the 2FA code sent to identity 'checkout-bot'")

Smolagents (Hugging Face)

Native Tool subclasses for Hugging Face smolagents:

python

from smolagents import CodeAgent, HfApiModel
from sentryrelay import SentryRelay
from sentryrelay.integrations.smolagents import create_smolagents_tools

relay = SentryRelay()
tools = create_smolagents_tools(relay)

agent = CodeAgent(
    tools=tools,
    model=HfApiModel(),
)

agent.run("Fetch the latest TOTP authenticator code for identity 'checkout-bot' with secret 'github_ops'")

LangGraph & LangChain

LangChain tool definitions with typed Pydantic parameter schemas:

python

from langgraph.prebuilt import create_react_agent
from langchain_openai import ChatOpenAI
from sentryrelay import SentryRelay
from sentryrelay.integrations.langgraph import create_sentryrelay_tools

relay = SentryRelay()
tools = create_sentryrelay_tools(relay)
llm = ChatOpenAI(model="gpt-4o")

app = create_react_agent(llm, tools=tools)

CrewAI

Prebuilt BaseTool instances for multi-agent crews:

python

from crewai import Agent, Crew, Task
from sentryrelay import SentryRelay
from sentryrelay.integrations.crewai import create_crewai_tools

relay = SentryRelay()
tools = create_crewai_tools(relay)

auth_specialist = Agent(
    role="Authentication Specialist",
    goal="Handle 2FA and MFA verification challenges",
    backstory="Expert at retrieving inbound OTP codes and managing TOTP credentials",
    tools=tools,
)

Production Best Practices for Autonomous Agents

Bound your timeout: Always set an explicit timeout_seconds (between 15 and 45s) to prevent agents from blocking indefinitely if an upstream service fails to send a code.
Use sender filtering: Pass sender="stripe.com" to filter out irrelevant emails and ensure the agent only claims the exact code intended for the active login challenge.
Rely on TOTP Rollover: Never manually calculate remaining seconds. Sentry Relay's auto_wait_fresh_window automatically delays if ≤4s remain in the window so the browser agent never types an expired token.