Recipes

Guard http requests

For an agent that fetches URLs, in 17 frameworks.

Last reviewed 2026-10-02

The cloud metadata address, the classic server-side request forgery target, is refused. For everything else, an egress allowlist of the hosts the agent may reach is the stronger control.

The rule

{
  "name": "http-guard",
  "effect": "deny",
  "priority": 10,
  "match_path": "/tools/http_get",
  "conditions": [
    {
      "field": "url",
      "op": "contains",
      "value": "169.254.169.254"
    }
  ]
}

Measured on production (2026-10-02, rule tester): the risky call below matches this rule and the safe one does not.

Risky	`{"url": "http://169.254.169.254/latest/meta-data/"}`
Safe	`{"url": "https://api.github.com/repos/acme/app"}`

Pick your framework

Framework	Tool name CAIN sees	Plugs into
Python (any function)	`http_get`	the function that acts
LangChain	`http_get`	your agent's tool list
LangGraph	`http_get`	the tools your graph's nodes call
OpenAI Agents SDK	`http_get`	function tools
CrewAI	`http_get`	CrewAI's before-tool-call hooks
LlamaIndex	`http_get`	FunctionTool
Pydantic AI	`http_get`	an agent capability
AutoGen (agentchat)	`http_get`	the callables you give AssistantAgent
Google ADK	`http_get`	the functions in an agent's tools
Claude Agent SDK	`mcp__ops__http_get`	PreToolUse hooks
MCP client (Python)	`http_get`	any MCP client session
OpenTelemetry	`http_get`	your existing tracing
Any other framework	`http_get`	any Python callable
Claude Code hooks	`WebFetch`	Claude Code's PreToolUse hook
Cursor hooks	`mcp__fetch__fetch`	Cursor's permission hooks
TypeScript / JavaScript	`http_get`	any async function
HTTP (any language)	`http_get`	your own call site

Try CAIN-42 on your own agents

Create a free account and every new account starts with a 7-day trial of the full platform. Or try the sandbox first, with no account at all.

Create a free account → · Try the sandbox · See the whole ecosystem