Skip to content

Claude ADK Integration

Wrap Claude ADK tools so every tool invocation passes through SafeAI's policy engine -- secret detection, PII filtering, tool contracts, and audit logging are enforced transparently.

Auto-generate Claude ADK integration

Let the intelligence layer generate SafeAI adapters for your Claude tools: 

safeai intelligence integrate --target claude-adk --path . --apply

Install

uv pip install safeai anthropic

Quick Start

from safeai import SafeAI

ai = SafeAI()
adapter = ai.claude_adk_adapter()
safe_tool = adapter.wrap_tool("search", search_tool, agent_id="claude-agent")

Detailed Usage

Creating the Adapter

from safeai import SafeAI

ai = SafeAI.from_config("safeai.yaml")
adapter = ai.claude_adk_adapter()   # returns SafeAIClaudeADKAdapter

You can also import the adapter directly:

from safeai.middleware.claude_adk import SafeAIClaudeADKAdapter

Wrapping Tools

def get_weather(location: str) -> str:
    """Get current weather for a location."""
    return f"Weather in {location}: 72F, sunny"

# Wrap with SafeAI
safe_weather = adapter.wrap_tool(
    name="get_weather",
    tool=get_weather,
    agent_id="assistant",
)

Request and response interception

The adapter intercepts both the request (tool input) and the response (tool output). Inputs are scanned before the tool executes; outputs are guarded before they are returned to the Claude model.

Wrapping Multiple Tools

tools = {
    "get_weather": get_weather,
    "search_docs": search_docs,
    "run_query": run_query,
}

safe_tools = {
    name: adapter.wrap_tool(name, fn, agent_id="assistant")
    for name, fn in tools.items()
}

Full Example

import anthropic
from safeai import SafeAI

# 1. SafeAI setup
ai = SafeAI.from_config("safeai.yaml")
adapter = ai.claude_adk_adapter()

# 2. Define tools
def search_database(query: str) -> str:
    """Search the internal database."""
    return f"Found 3 results for: {query}"

def send_email(to: str, subject: str, body: str) -> str:
    """Send an email."""
    return f"Email sent to {to}"

# 3. Wrap tools
safe_search = adapter.wrap_tool("search_database", search_database, agent_id="assistant")
safe_email = adapter.wrap_tool("send_email", send_email, agent_id="assistant")

# 4. Define tool schemas for Claude
tool_schemas = [
    {
        "name": "search_database",
        "description": "Search the internal database",
        "input_schema": {
            "type": "object",
            "properties": {"query": {"type": "string"}},
            "required": ["query"],
        },
    },
    {
        "name": "send_email",
        "description": "Send an email",
        "input_schema": {
            "type": "object",
            "properties": {
                "to": {"type": "string"},
                "subject": {"type": "string"},
                "body": {"type": "string"},
            },
            "required": ["to", "subject", "body"],
        },
    },
]

# 5. Use with Anthropic client
client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    tools=tool_schemas,
    messages=[{"role": "user", "content": "Search for Q4 revenue data"}],
)

# 6. Handle tool use -- SafeAI enforces policy here
for block in response.content:
    if block.type == "tool_use":
        if block.name == "search_database":
            result = safe_search(query=block.input["query"])
        elif block.name == "send_email":
            result = safe_email(**block.input)

Sensitive data is caught

If Claude tries to include credentials in an email body or leak PII through a search query, SafeAI blocks the tool call and logs the violation.

Configuration

# safeai.yaml
policy:
  default_action: block
  secret_detection:
    enabled: true
  pii_protection:
    enabled: true
    action: redact

tool_contracts:
  send_email:
    allowed_agents: ["assistant"]
    max_calls_per_minute: 5
    approval_required: true
  search_database:
    allowed_agents: ["assistant"]

audit:
  enabled: true
  log_inputs: true
  log_outputs: true

API Reference

Class Description
SafeAIClaudeADKAdapter Main adapter returned by ai.claude_adk_adapter()
adapter.wrap_tool() Wrap a single tool with policy enforcement

See API Reference - Middleware for full signatures.

Next Steps