The Network · as a tool for an assistant

Ask an assistant about the Network.
The same page, read by a model.

AlphaHub runs an MCP server: one address that turns the Network page into a set of tools for Claude, ChatGPT, Cursor or an agent of your own. The assistant reads the hub's numbers, the manual and an FAQ, and answers under the same caveats the page prints. Public, free, no account.

addresshttps://api.alphanode.tech/mcp

public data only

Exactly what the Network page shows to anyone: public ids, pairs, the hub's return, Sharpe, confidence and drawdown since hosting, the promised backtest, the price when on sale.

no formula, no holdings

The hub holds formulas sealed and shows them to nobody; what a strategy holds goes only to those who bought it. The assistant gets neither, nor an e-mail, a device or an error text.

reads, never writes

Nothing here hosts, lists, quotes or pays. Buying stays on the site, from your wallet, on your decision. The assistant describes and compares; it does not recommend buying.

Connect

One address, any client

The server speaks MCP over HTTP, without a login. Add the address where your assistant keeps its connectors, then ask in plain language.

CLAUDE

claude.ai and Claude Desktop

Settings → Connectors → Add custom connector: name it AlphaHub, paste the address, leave the OAuth fields empty, add. In a chat, switch the connector on in the tools menu. Custom connectors come with the Pro, Max and Team plans.

CLAUDE CODE

In the terminal

claude mcp add --transport http alphahub https://api.alphanode.tech/mcp
# then, inside a session: /mcp lists the tools

CHATGPT

As a connector

Settings → Connectors → Advanced → Developer mode → Create, the address, no authentication. The server has the search and fetch pair ChatGPT expects.
CURSOR · OTHERS
Any MCP client over HTTP
```
{"mcpServers": {"alphahub": {"url": "https://api.alphanode.tech/mcp"}}}
```
Your own agent connects the same way: the address as an HTTP MCP server, in the MCP connector of the Anthropic API or any MCP client library. Stateless: one JSON-RPC call per request, a JSON answer, no session to keep.

Try asking

What is on the Network right now, and which 4h strategies hold a confidence above 70%?

Explain strategy 3f9a1b2c4d5e: how long has the hub run it, and how does the track compare with what was promised?

What is on sale, at what price, and how did this node's earlier listings end?

How do I run a node in Docker, and will it mine again after a reboot?

What it answers

Nine tools, the manual and an FAQ

Every tool reads the same public data as the page, through the same cache; the model picks the one your question needs.

network_overviewThe Network at a glance: strategies, nodes, live and in profit, the median return, the split by bar and kind, how many are on sale, the equal-weight index and the growth.
network_strategiesThe list, filtered by words, bar, kind or on sale; sorted by confidence, return, Sharpe, drawdown, age or bars; paged.
network_strategyOne strategy opened: its numbers, the backtest its author's node promised, the cost and sizing assumptions, the hub's track, its page link.
network_nodeOne node: every strategy it hosts and the sales it ended, with the numbers each showed when it stopped selling.
network_howHow the Network works, in its own words, and the whole AlphaNode manual by entry number or by words.
faqThe questions people ask, answered from the manual and pointing at its entries: getting started, mining, reading the leaderboard, Docker, hosting, selling and buying.
alphanode_pricingWhat AlphaNode costs, and where to buy and download it.
search · fetchThe connector pair ChatGPT expects, over the same strategies, FAQ and manual.

Also as resources an assistant can pin to a conversation: the overview, the FAQ, the whole manual, one strategy or one manual entry by address; and two ready prompts, scout (find by criteria, describe honestly) and strategy (explain one). The full list is in the connector's own tool view.

What the assistant is told

The page's caveats, as rules

The server hands every model the same instructions before the first question.

simulated

Not a live account

Every return is simulated on market data with fees and funding. The assistant must say so and must never present a number as money someone made.

the track counts

The promise does not

The backtest an author's node measured before hosting cannot be verified by the hub and reads better than it will live. Sharpe appears from 30 bars; a confidence near 50% means nothing yet.

no advice, no money

Describe, compare, link

Formulas are sealed and not to be guessed at. The assistant cannot quote, buy, sell or pay and must not recommend buying: it gives the strategy's page, and the decision is yours.

Plain answers

Questions about the connector

Is it free?

Yes. The server runs no model of its own; it answers with data. The model runs on your side, on your assistant's plan or your API key.

Does it need an account?

No, and it has nothing personal to show: no favourites, no bought signals, no hosted books of yours. Account tools under a scoped key are the next step; buying will stay on the site either way.

Can it trade or buy for me?

No. It reads. Orders are not part of AlphaNode at all, and a purchase is a payment from your wallet on the Network page.

Which clients work?

Any MCP client that speaks the HTTP transport: Claude (web, desktop, Code), ChatGPT connectors, Cursor, Windsurf, Gemini CLI, the Anthropic API's MCP connector, the MCP client libraries. There is no event stream and no session; a client that insists on either falls back to plain requests.

Are there limits?

300 calls a minute per address, answers cached for 30 seconds like the page's own. Enough for any conversation; a scanner meets a 429.

Can I see the numbers it sees?

Open the Network page: it is the same data, drawn. The manual's Part 10 says what each column means.

Ask an assistant about the Network.The same page, read by a model.