Agent frameworks

Criora’s MCP server at https://criora.com/mcp works in any agent framework that speaks the Model Context Protocol over Streamable HTTP. It needs no account and no key. Each example below gives an agent every Criora tool, and each was run against the live server with the library versions it names. Data API and AI agents lists the tools and what each one answers.

LangChain

pip install langchain langchain-anthropic langchain-mcp-adapters

import asyncio

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

CRIORA = {"transport": "streamable_http", "url": "https://criora.com/mcp"}


async def main():
    tools = await MultiServerMCPClient({"criora": CRIORA}).get_tools()
    agent = create_agent("anthropic:claude-sonnet-5-5", tools)
    question = "Is it safe to travel to Manila next week?"
    answer = await agent.ainvoke({"messages": question})
    print(answer["messages"][-1].content)


asyncio.run(main())

Run with langchain 1.4 and langchain-mcp-adapters 0.3. Any chat model LangChain supports can take the Claude model’s place.

LlamaIndex

pip install llama-index-core llama-index-tools-mcp llama-index-llms-anthropic

import asyncio

from llama_index.core.agent.workflow import FunctionAgent
from llama_index.llms.anthropic import Anthropic
from llama_index.tools.mcp import BasicMCPClient, McpToolSpec


async def main():
    criora = McpToolSpec(client=BasicMCPClient("https://criora.com/mcp"))
    tools = await criora.to_tool_list_async()
    agent = FunctionAgent(tools=tools, llm=Anthropic(model="claude-sonnet-5-5"))
    question = "Which countries face the most dangerous weather this week?"
    print(await agent.run(question))


asyncio.run(main())

Run with llama-index-core 0.14 and llama-index-tools-mcp 0.4.

OpenAI Agents SDK

pip install openai-agents

import asyncio

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


async def main():
    params = {"url": "https://criora.com/mcp"}
    async with MCPServerStreamableHttp(name="Criora", params=params) as criora:
        agent = Agent(
            name="Risk analyst",
            instructions="Answer with Criora's data and name its sources.",
            mcp_servers=[criora],
        )
        question = "How exposed is Rotterdam to river flooding by 2050?"
        result = await Runner.run(agent, question)
        print(result.final_output)


asyncio.run(main())

Run with openai-agents 0.20. To let OpenAI’s Responses API call the server itself, with no MCP client in your code, give the agent HostedMCPTool(tool_config={"type": "mcp", "server_label": "criora", "server_url": "https://criora.com/mcp", "require_approval": "never"}) in tools instead.

n8n

1. Add an AI Agent node and connect a chat model to it.
2. Under Tool, add an MCP Client Tool.
3. Set Endpoint to https://criora.com/mcp, Server Transport to HTTP Streamable, Authentication to None and Tools to Include to All.

Good to know

• Answers are concise by default: the fields that answer the question. Pass response_format: "detailed" to a tool for every field, such as each report’s provenance or a warning’s full text.
• Every answer names its sources, and a hazard answer gives each source’s licence and attribution in credit. When you publish what an agent found, credit Criora and the source it names, as Data API and AI agents sets out under sources and attribution.
• The server is read-only and free to use under the Terms of Service.