An agent that can only use the tools its author wired in at build time is limited to what its author thought of. This endpoint lets one ask, at runtime, what exists for a task, across this index and every other public ARD registry at once.

One endpoint for all of it: https://neuronto.com/mcp. No account, no key.

from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client
from langchain_mcp_adapters.tools import load_mcp_tools
from langchain.agents import create_agent

async with streamablehttp_client("https://neuronto.com/mcp") as (read, write, _):
    async with ClientSession(read, write) as session:
        await session.initialize()
        tools = await load_mcp_tools(session)
        agent = create_agent("openai:gpt-4.1", tools)

Every tool the registry exposes becomes a LangChain tool, so find_resource and find_tool are callable by the agent itself.

from agents import Agent
from agents.mcp import MCPServerStreamableHttp

async with MCPServerStreamableHttp(
    params={"url": "https://neuronto.com/mcp"},
    name="Neuronto ARD Registry",
    cache_tools_list=True,
) as server:
    agent = Agent(name="Assistant", mcp_servers=[server])

cache_tools_list=True is worth setting: the tool list here changes rarely and the round trip is not free.

import { experimental_createMCPClient as createMCPClient } from 'ai';

const client = await createMCPClient({
  transport: { type: 'http', url: 'https://neuronto.com/mcp' },
});

const tools = await client.tools();
// ...use tools, then release the connection
await client.close();

Close the client when you are done; it holds a connection open. The SDK also accepts MCP's official StreamableHTTPClientTransport if you already use it.

Discovery is a plain HTTP call, so a framework with no MCP support can still use it. The npm package wraps it with no dependencies:

import { findResource } from 'neuronto';

const { results } = await findResource('read a PDF and extract tables');
for (const r of results) console.log(r.displayName, r.url, r.score);

Python: pip install ard-publish for the publishing side, or POST https://neuronto.com/search directly with {"query": {"text": "..."}}. Both are documented at /api-docs.

LlamaIndex, CrewAI, Mastra, Pydantic AI, Google ADK, Microsoft Agent Framework and Strands all connect to remote MCP servers. Their exact call signatures are not reproduced here because they were not verified against source at the time of writing, and a snippet that looks authoritative and is wrong wastes more of your time than none. Point whatever the framework calls a streamable HTTP MCP server at:

https://neuronto.com/mcp

If a snippet here is wrong or one is missing, tell us and it gets fixed rather than argued about.

Ranked matches with the endpoint to connect to, which registries carried each one, and what was verified by fetching it. score is semantic relevance only and is never a trust or safety rating. Editors and CLIs are at /connect.