RingMaster in your AI

Last updated 2026-09-24

You already ask an AI about the market. It has no evening ranking, no published book and no record to check itself against. This connector hands it ours: a read only door into the desk that your own AI client can open. You paste one address, sign in, press Allow, and ask.

At a glance

What it reads. The desk's view of one name, the published books, the evening ranking, one name's card and what the option chain shows. Also its daily price bars, the market reading, the sector view, the release calendar, public filings, the paper record and the retired research.
What the view is. Ask whether a name is bullish or bearish and your AI gets the desk's answer, not its own. Which side, bullish or bearish, over the near horizons and the long ones, with the levels the desk watches.
What it cannot do. It cannot place an order, change anything, or read anything your account does not already reach. Every tool is a read.
What it costs. Nothing extra on any plan.
Key difference. Your AI stops guessing from a chart. It quotes a published figure with the date and the receipt attached.
Where the arithmetic starts. The connector serves the same daily bars the desk computes its own figures from. Your AI and the desk work off one series.
Every figure is paper trading and hypothetical. The connector says so on every answer, so your AI can repeat it.

You need a RingMaster account. There is no anonymous access to this door, and there never will be. Sign up first, then come back here.

Open Settings from the cogwheel, then Your AI. Copy the address at the top of the panel with the Copy button.

That address is your desk with /mcp on the end. The examples below use the public one.

Then add it in your own client. Each one takes about a minute.

Claude, on the web, desktop or phone

Open Settings, then Connectors. Choose Add custom connector.

Paste the address and press Continue. Sign in to RingMaster, then press Allow.

Claude Code

Run this in a terminal:

claude mcp add --transport http ringmaster https://ringmaster.oheyinc.com/mcp

Then type /mcp inside Claude Code and sign in.

Cursor

Open Settings, then MCP. Choose Add and paste the address.

ChatGPT

Turn on Developer mode first. Open Settings, then Connectors, then Advanced.

Then choose Create, paste the address, pick OAuth and sign in.

Developer mode needs a paid ChatGPT plan. That is OpenAI's rule, not ours.

Anything else that speaks the protocol

The transport is streamable HTTP. Point the client at the address.

It reads the sign-in details from the desk and asks you to allow it.

What you are agreeing to

The screen names the client that asked and lists what it can read. It reads and never writes.

It cannot place an order, change anything on the desk, or read past what your account already shows you.

You can take it back at any time. Settings, Your AI, Connected apps, Remove. Every token under it stops working at once.

Step two: a key, for a client that asks for a header

Some clients want a key in a header instead of a sign-in. Settings, Your AI, at the bottom of the panel.

Give the key a name you will recognise, such as Laptop or Phone, and press Mint a key.

The key is shown once. Copy it before you leave the panel. The desk keeps a fingerprint of it and the last four characters, never the key itself.

You can hold up to eight keys at a time. One per device is the tidy way to do it. Withdrawing a lost phone's key then leaves your laptop working.

Send it as an Authorization: Bearer header. A configuration file looks like this.

{
  "mcpServers": {
    "ringmaster": {
      "url": "https://ringmaster.oheyinc.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY_HERE" }
    }
  }
}

What each tool gives your AI

The Delayed for Free column says whether a free account reads the previous published trading day. Every answer's stamp names the trading day it came from. Your AI never has to guess.

Tool	What comes back	Delayed for Free
Trade Cards	Tonight's signals, one row a name, each with a probability beside the same measurement for any stock, an option strategy and a limit. It also serves the hot, popular, favourites, Leaders and theme lists.	Yes
Tonight's model book	The model book published after each close: five names bullish, five bearish, equal weight. The current model ranks it on its twenty trading day rank from 25 Sep 2026; the earlier model ranked it before that. Each answer names its model and the exit rule its paper record is scored under.	Yes
The evening ranking	About a thousand names ranked, filterable by side, by sector and by how many you want. Each row carries the side and the measured outlook: a probability beside the same figure for any stock, with a confidence. It also carries the ranks at 10, 20 and 45 trading days, and how many trading days its signal state has run. Last night's options reading is joined in.	Yes
The desk's view of one name	Which way the desk leans on one name, with the measured outlook beside it. It names two levels either side of the last close, each a price the name already turned at, with its date. The desk publishes no price target of its own.	Yes
One name's card	Where a ticker ranks at 10, 20 and 45 trading days, its measured outlook beside any stock, and how long its signal state has run. It also carries which books hold it, any flags, a dated chart read from the desk's own bars, and last night's option chain reading.	Yes
The trading day reading	The volatility label, the print it came from and how old it is. Also how many names were ranked, the coverage screen, and how many are bullish or bearish.	Yes
Sector by sector	Each sector's tilt, how many names sit each way, and how many the books hold. Also its top ranked name, and how far the tilt moved since the last trading day.	Yes
Replay a past trading day's book	What the books held on a past evening, plus the list of trading days on file, so you can replay the record yourself. Each trading day carries the model book as published and the model that ranked it.	Yes, by one trading day
One name across trading days	A ticker's place in the ranking at each time frame on each recent evening, oldest first. When it moved far enough to make the desk's own list, how many places it shifted comes back too.	Yes
Price history	Up to five years of daily bars for one name, split adjusted, from the desk's own feed. It is the same series the ranks and chart reads come from.	No
Names side by side	Up to six tickers as one table: side, the measured outlook, and the ranks at 10, 20 and 45 trading days. Also how long each signal state has run, which books hold each, and a one line option chain reading.	Yes
What the options market shows	Open interest on each side, implied volatility across expiries, the shape of the skew, and open dealer gamma for a one percent move.	No
The options screen	The whole covered universe as one table, sortable by open interest, implied volatility, puts over calls, either overnight change, or open gamma. Each row also says where implied volatility sits against its own history, and what the spread costs to cross.	No
What moved in one chain overnight	One name's chain across the two most recent captures, each figure stated as a pair rather than a difference.	No
What changed hands	How many contracts on one name traded in the captured trading day, by side and by expiry, with open interest printed beside them. The busiest single contracts come back too.	No
Implied volatility history	One name's long weekly at the money implied volatility history, or its recent gamma and delta replay, one trading day at a time.	No
The desk column for one name	The desk analyst's published note for a ticker, with the date of the set it was written in.	Free accounts read the opening paragraph
What is announced	The scheduled economic releases and the covered companies due to report, inside a date range you choose.	No
Market Radar	One Market Radar section: report dates, filings, short sale volume, seasonality, market mood, breadth, the macro tape, grade changes, insider trades or headlines.	No
Public filings for one name	Congressional and insider filings on a ticker as filed, with who filed, when, and a link to the original.	No
What the desk covers	The coverage list and the liquidity screen behind it, so a script knows what is ranked and what is not.	Yes
The public paper record	The headline cells the record page leads with, the exit rules table, the competitor comparison, and the forward ledger.	No
The retired research	Search what the firm tried and dropped. Each entry names the study, how many settings it tested, what it measured and the verdict.	No
What this desk served you	Your own rows in the delivery trail, with the hashes over them, so you can audit what you were served.	No
Find something on this desk	A search across the guides, the covered names, the published books, the record and the retired research.	No
Read one thing in full	The full text of anything the search found, by its id. A name's id answers with the desk's view of that name.	No

Ten questions to ask once connected

Each of these is a read. None of them asks your AI to predict anything.

What are tonight's signals on Trade Cards, and what is in the model book?
Under what rule does the published record score that book?
What is the volatility label this trading day, and how many names were ranked?
Which two sectors are tilted furthest each way, and how far did each move?
Where does NVDA sit at each time frame, and which books hold it?
Compare NVDA, AMD and AVGO side by side on the desk's own columns.
What did the book hold on 2026-08-14?
What does the option chain show on TSLA, and what changed overnight?
Which covered companies report in the next seven days?
Show me my own delivery rows, and what the hash chain does not prove.

Ask your AI to quote the stamp when it quotes a figure. Good clients do it without being asked once they have seen one.

For a script

Three requests are the whole protocol. Put your key in the header.

Start with the handshake:

curl -s https://ringmaster.oheyinc.com/mcp \
  -H "Authorization: Bearer $YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize",
       "params":{"protocolVersion":"2026-07-28","capabilities":{},
                 "clientInfo":{"name":"my-script","version":"1"}}}'

List what the desk serves:

curl -s https://ringmaster.oheyinc.com/mcp \
  -H "Authorization: Bearer $YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'

Then call one:

curl -s https://ringmaster.oheyinc.com/mcp \
  -H "Authorization: Bearer $YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call",
       "params":{"name":"get_tonight_book","arguments":{"book":"basic"}}}'

Every answer carries a data block and a stamp block. The stamp names the trading day and when the desk published it. It also carries a receipt over the exact figures, and the page where a person reads the same thing.

Your account may make 600 requests a day. Every connection you hold shares that one limit. It resets at midnight UTC.

Six hundred is a lot of reading. A nightly routine that pulls the books, the ranking and a card for twenty names spends about twenty three of them.

A second key does not buy a second 600. If one script runs away it spends the day for every connection, so the meter in Settings is worth a look.

You can see what you have spent. Open Settings, then Your AI. It shows today's count for each connection, the last thirty days, and which apps did the asking. The phone shows the same three answers under More, then Your AI.

For ChatGPT

ChatGPT reaches this desk through two tools called search and fetch. That is OpenAI's shape for a connector, and every connector in ChatGPT works this way.

search takes words and answers with a list of things it found. Each result has an id, a title and a page address.

fetch takes one of those ids and answers with the full text.

You do not have to use them. Ask your question in everyday words and ChatGPT will pick. The other tools are there too, and a specific question reaches a specific tool.

Developer mode is where custom connectors live in ChatGPT. It sits under Settings, Connectors, Advanced, and it needs a paid ChatGPT plan.

Ready made prompts and the guides

Some clients show a menu of prompts a connector supplies. This one offers four: a morning brief, a walk through one name, what changed since the last trading day, and a check of the record.

Each one tells your AI which tools to call, in what order, and to quote the stamp beside any figure it repeats.

The desk's guides are offered as documents too, at addresses beginning ringmaster://guide/. A client that supports resources can attach one and read how a pane works before it reports a number from it.

A skill bundle is offered the same way, at addresses beginning ringmaster://skill/. Each one is a short file that teaches a client which tools to call, in what order, for one kind of question. The same four are published for download at oheyinc.com/ai.html.

What a free account sees

A free account reads the previous published trading day on the books and the ranking. The stamp says which trading day you got, every time.

The public record is never delayed. Neither is the option chain reading, the release calendar, the public filings or the retired research.

Things worth knowing

The key acts as you. What it may read is checked against your account on every request, not stamped on the key. Changing your plan changes what it reads, at once.
Withdraw a key you have lost. Settings, Your AI, Withdraw. It stops working immediately, and the row stays on your list with the date.
No writes, ever. There is no tool that places an order, edits a list or changes a setting. The desk publishes, and this door is a page.
It reads, it does not think. The connector hands your AI figures. What your AI says about them is your AI's, and it is worth reading with the same care you would give any other opinion.