Agent SDK snippets

Server

https://cve.openkrill.app/mcp

Transport

Streamable HTTP

Key

None

pip install openai-agents

import asyncio
from agents import Agent, Runner
from agents.mcp import MCPServerStreamableHttp


async def main():
    async with MCPServerStreamableHttp(
        name="cve",
        params={"url": "https://cve.openkrill.app/mcp"},
        cache_tools_list=True,
    ) as server:
        agent = Agent(
            name="Security assistant",
            instructions="Answer using the available tools.",
            mcp_servers=[server],
        )
        result = await Runner.run(agent, "How bad is CVE-2021-44228?")
        print(result.final_output)


asyncio.run(main())

pip install langchain langchain-mcp-adapters langchain-openai

import asyncio
from langchain.agents import create_agent
from langchain_mcp_adapters.client import MultiServerMCPClient


async def main():
    client = MultiServerMCPClient({
        "cve": {
            "url": "https://cve.openkrill.app/mcp",
            "transport": "streamable_http",
        }
    })
    tools = await client.get_tools()
    agent = create_agent("openai:gpt-4.1", tools)
    result = await agent.ainvoke(
        {"messages": [{"role": "user", "content": "How bad is CVE-2021-44228?"}]}
    )
    print(result["messages"][-1].content)


asyncio.run(main())

npm install @openai/agents zod

import { Agent, MCPServerStreamableHttp, run } from "@openai/agents";

const server = new MCPServerStreamableHttp({
  name: "cve",
  url: "https://cve.openkrill.app/mcp",
});
await server.connect();
try {
  const agent = new Agent({
    name: "Security assistant",
    instructions: "Answer using the available tools.",
    mcpServers: [server],
  });
  const result = await run(agent, "How bad is CVE-2021-44228?");
  console.log(result.finalOutput);
} finally {
  await server.close();
}

Check the connection without a model

To test the server before you spend a model call, list its tools and call one directly. This is how each snippet above was checked: the call returns the same answer the agent would read. In Python, inside main():

async with MCPServerStreamableHttp(
    name="cve", params={"url": "https://cve.openkrill.app/mcp"}
) as server:
    print([tool.name for tool in await server.list_tools()])
    result = await server.call_tool(
        "check_cve", {"cve": "CVE-2021-44228"}
    )
    print(result.is_error, result.content[0].text[:200])

The tool list should include check_cve, and the first value printed should be False. LangChain gives the same check with await client.get_tools() and await tool.ainvoke({...}) on the check_cve tool.

Limits

Each server takes 60 calls a minute from one network, and a server has a shared daily cap. A refused call comes back as a tool error that starts with rate_limited or daily_limit, so the agent reads it as an error, not as an empty answer. The other answers are in If it does not work. For higher limits, ask support@openkrill.app for a key.

Other servers

Every server uses the same pattern: change the name and the URL. Install steps for ChatGPT, Claude, Cursor, VS Code and Pi are in the MCP install guides.

• Company Jobs Finder: https://jobs.openkrill.app/mcp
• Startup Name Check: https://names.openkrill.app/mcp
• Recall Check: https://recalls.openkrill.app/mcp
• Citation Checker: https://citations.openkrill.app/mcp
• Store Price Tracker: https://stores.openkrill.app/mcp
• Business Days & Workday Calc: https://businessdays.openkrill.app/mcp
• Tariff & HS Code Duty Finder: https://tariffs.openkrill.app/mcp
• EU Business & VAT Checker: https://eubusiness.openkrill.app/mcp
• Regulation Reader: https://rules.openkrill.app/mcp
• Site SEO & AI Crawler Audit: https://sitecheck.openkrill.app/mcp
• Travel Check: https://travel.openkrill.app/mcp
• Food Facts: https://food.openkrill.app/mcp
• Car Check: https://cars.openkrill.app/mcp
• Document Tools: https://docs.openkrill.app/mcp
• Tax Numbers: https://tax.openkrill.app/mcp
• Catalog QA Desk: https://catalogqa.openkrill.app/mcp
• Dev Facts Check: https://devfacts.openkrill.app/mcp
• CVE Risk Check: https://cve.openkrill.app/mcp
• Domain Mail Check: https://mailcheck.openkrill.app/mcp
• SEO Audit: https://seoaudit.openkrill.app/mcp
• Page Fetch: https://pagefetch.openkrill.app/mcp
• Post Stats: https://poststats.openkrill.app/mcp
• IndexNow Submit: https://indexnow.openkrill.app/mcp

Source: runs of each snippet against https://cve.openkrill.app/mcp on 2026-10-21, up to the model call. As of 2026-10-21.