Skip to content

Getting Started

Get AgentShield running in under 5 minutes.

Installation

# Core library (zero dependencies)
pip install agentshield-fw

Optional Extras

pip install "agentshield-fw[mcp]"          # MCP server support
pip install "agentshield-fw[langchain]"    # LangChain adapter
pip install "agentshield-fw[crewai]"       # CrewAI adapter
pip install "agentshield-fw[openai]"       # OpenAI Agents SDK
pip install "agentshield-fw[dashboard]"    # Real-time dashboard
pip install "agentshield-fw[hitl]"         # Human-in-the-loop gateway
pip install "agentshield-fw[otel]"         # OpenTelemetry export
pip install "agentshield-fw[all]"          # Everything
pip install -e ".[dev]"                    # Development tools

Protect Any Python Function

The simplest integration — wrap any function with @shield.protect:

import agentshield

shield = agentshield.Shield()

@shield.protect
def execute_sql(query: str) -> str:
    return db.execute(query)

@shield.protect
def read_file(path: str) -> str:
    return open(path).read()

# Safe calls pass through transparently
result = execute_sql("SELECT * FROM users WHERE id = 1")

# Dangerous calls are blocked
try:
    execute_sql("DROP TABLE users")
except agentshield.ToolCallBlocked as e:
    print(f"Blocked: {e.response.reason}")
    print(f"Rule: {e.response.rule_name}")
    print(f"OWASP: {e.response.owasp_id}")

Protect an MCP Server

from mcp.server import Server
from agentshield.adapters.mcp import shield_mcp_server

server = Server("my-server")
shield_mcp_server(server)

# All MCP tool calls now pass through AgentShield

Protect a LangChain Agent

from langchain.tools import Tool
from agentshield.adapters.langchain import ShieldedToolkit

tools = [sql_tool, file_tool, search_tool]
shielded = ShieldedToolkit(tools)

agent = create_react_agent(llm, shielded.tools)

Protect a CrewAI Crew

from crewai import Crew
from agentshield.adapters.crewai import shield_crew

crew = Crew(agents=[...], tasks=[...])
shield_crew(crew)

Protect an OpenAI Agent

from agents import Agent
from agentshield.adapters.openai_sdk import shield_agent

agent = Agent(name="assistant", tools=[...])
shield_agent(agent)

Configuration

Create agentshield.yaml in your project root for custom configuration:

mode: enforce

rules:
  destructive_sql:
    enabled: true
  rate_limiter:
    enabled: true
    max_calls: 100
    window_seconds: 60

audit:
  file: shield.jsonl
  hash_chain: true

See the Configuration Reference for all options.

CLI Commands

# Verify audit log integrity
agentshield verify shield.jsonl

# Show statistics from audit log
agentshield stats shield.jsonl

# Export audit log to CSV
agentshield export shield.jsonl --format csv

# Start the dashboard
agentshield serve --port 9090

Next Steps