AI agent for trading: 40 prompts for ChatGPT and Claude
You can now trade on Upscale in an ordinary chat with an AI. An AI agent — ChatGPT, Claude, Claude Code or Cursor — connects to your account, and from then on you simply write to it in plain text: “how much room is left before the daily limit?”, “buy $100 of BTC with a stop 5% below”, “review my trading this phase”. The agent pulls account and market data itself, does the maths and places orders, while the decisions stay with you. It all works over MCP, an open standard for connecting AI assistants to external services. Below you will find what MCP is, how to connect an agent and 40 ready-made prompts you can copy into the chat as a cheat sheet: from checking your challenge limits to trading in a single sentence and reviewing your own stats.
What MCP is and how an AI agent differs from a bot
MCP (Model Context Protocol) is an open protocol that lets AI assistants connect to external services: pull data from them and call their functions. The standard was introduced by Anthropic in November 2024, OpenAI and Google adopted it in 2025, and in December 2025 the protocol was handed over to a foundation under the Linux Foundation. Put simply, MCP is an adapter between a chatbot and a platform: the AI understands your request, and the MCP server turns it into actions on your account.
Upscale has its own MCP server. An agent connected to it sees your accounts and the platform's markets, and can place, modify and cancel orders. Here is how this compares with other ways to trade:
| Terminal | Bot via API | AI agent via MCP | |
|---|---|---|---|
| Who makes the decision | The trader | An algorithm following set rules | The trader; the agent calculates and executes |
| How commands are given | Buttons in the interface | Code | Plain text |
| Coding required | No | Yes | No |
| Strength | Charts and a visual overview | Speed and non-stop operation | Analysing account data and acting in one sentence |
The agent does not replace a terminal with charts, and it does not trade on its own: it responds to your requests. Its strength lies elsewhere. To check the room left before the drawdown limit on three accounts, compare them and close a position, the terminal takes several screens, while the agent needs one message. If you want fully automated trading on strict rules, that is what API trading is for. And we covered the different levels at which AI can help a trader in our guide to AI trading.
How to connect an agent to Upscale
The agent only sees accounts with the API Trading option. It can be added to accounts of $10,000 and above before they move to the funded stage, and on a demo account it is enabled when the account is created. There are two ways to connect an agent:
- By server address. Open the API and MCP settings, go to the MCP tab, click "Connect" and choose your agent — ChatGPT, Claude, Claude Code and Cursor each have step-by-step instructions. After you add the address
, the agent opens an "Allow access" page showing the app name and its permissions: viewing data and trading. Once you confirm, there is no key to enter.https://mcp.upscale.trade - By API key. Create a key in the API and MCP settings and add it, together with the server address, to the agent's configuration. There is no need to paste the key into chat messages.
Some capabilities depend on the app itself. For example, in ChatGPT, connecting external MCP servers and actions that change data are not available on every plan — see the current terms in the OpenAI help centre. In Claude, you can add your own connector even on the free plan, but only one — more in the Claude help centre. Cursor asks for permission before every MCP tool call by default.
To check the connection, send the agent your first prompt:
💬 “Show my Upscale accounts”
Agents connected by server address appear on the MCP tab with the date they were connected and last used. Any of them can be disconnected with one button, and the agent loses access to your accounts immediately. Step-by-step instructions are on the API Trading page.
Connecting takes a few minutes, and it is best to start on a demo account: quotes and conditions there are real, while a mistake in a prompt costs nothing.
Account and challenge rules
Along with account data, the agent receives your challenge limits from Upscale: phase and status, balance, equity (balance including open positions), start-of-day equity, peak equity, the daily, maximum and trailing drawdown limits, the profit target and the number of profitable days. So it answers questions about the rules with your own numbers rather than general statements.
Room left before the limits
💬 “How are my accounts doing? How much room is left before the daily drawdown limit?”
The agent calculates the room left before each limit and what remains to the profit target. We explained the types of limits and how they are calculated in our guide to prop firm drawdown types.
The 30% rule and the withdrawable amount
💬 “Am I meeting the 30% rule? How much can I withdraw right now?”
The agent sees whether the consistency rule applies to the account and whether it is currently met, as well as the amount available to withdraw. The 30% rule means your most profitable day must not account for 30% or more of total profit. The withdrawal itself can only be requested in the Upscale interface.
Payout readiness
💬 “What do I need to qualify for a payout this week?”
For this prompt the agent puts together, in one answer, the number of profitable days, whether the 30% rule is met, the withdrawable amount and the room left before the limits. To learn how payouts work, read our guide to prop firm payouts.
Why an account failed
💬 “Why was I knocked out, and what happened to my equity before that?”
The agent sees the reason for the failure — daily, maximum or trailing drawdown — the date, minute-by-minute equity for the 5 days before that moment and the closed trades of the current phase. You can ask for a plan straight away:
💬 “Break down what led to the failure and suggest three rules for my next challenge”
The full review method is in our article on how to analyse a failed prop challenge in 7 steps.
Account history and comparison
💬 “What happened on my accounts over the last month?”
💬 “Compare my accounts: where am I closest to the profit target?”
In the history the agent sees challenge purchases, passed and failed phases, triggers of the daily drawdown protection, withdrawal requests and their outcome, option purchases, and demo accounts created and closed. It retrieves all accounts with the API Trading option in a single request, so a comparison takes one message.
Instead of working out the room left before the limits on each account by hand, you get ready-made figures and see straight away where the risk is closest.
Trading in plain words
The agent places orders, manages open positions and closes them. Orders it places appear in the account's list of active orders, just as with manual trading. If an order is invalid — for example, the trigger price is on the wrong side of the market or the leverage is above what the account's phase allows — the platform rejects it and gives the reason.
A trade in one sentence
💬 “Buy $100 of BTC with 3x leverage, stop 5% below, take profit 5% above”
The agent opens a market position and sets the stop loss and take profit straight away.
Pending entry
💬 “Place a limit buy on BTC 3% below the market with a stop and take profit”
💬 “Go long BTC if the price rises 3%”
💬 “Go short BTC if the price falls 3%”
Limit, stop-market and stop-limit orders are available for entries. A stop loss and take profit can be attached to a limit order straight away.
Managing a position
💬 “Move the BTC stop to 80,000”
💬 “Move the take profit to 4% above the current price”
💬 “Set a 3% trailing stop”
💬 “Close half the position”
💬 “Add $20 of margin”
A trailing stop is a stop loss that follows the price at a set distance. Margin can be removed as well as added: “remove $10 of margin”.
Cancelling and “close everything”
💬 “Cancel all orders on the account”
💬 “Close everything on the account”
The second command cancels all orders and closes all positions in one action. You can also cancel a single order (“cancel the BTC limit order”) or close positions only (“close all positions”).
Familiar terminal actions come down to a single sentence, and when the market moves sharply, an emergency exit takes seconds.
Trading review and market overview
Your trading stats
💬 “Review my trading: win rate, where I lose, how long I hold losing trades compared with winning ones”
For each account the agent sees the number of trades, winning and losing trades, win rate (the share of profitable trades), total profit, fees and funding, average and largest profit and loss, the risk-to-reward ratio (risk/reward), maximum drawdown, position holding time, average leverage in winning and losing trades, and the most-traded market. These data lead to precise questions to ask yourself:
💬 “Where do I make more — on longs or shorts?”
💬 “Do I use higher leverage on losing trades than on winning ones?”
💬 “Am I holding losers longer than winners?”
Why a trade shows zero profit
💬 “Why is the profit on this trade 0?”
The agent sees whether the 60-second adjustment was applied to the trade and what the profit was before it. On accounts with the API Trading option, profit on a position closed less than 60 seconds after it was opened or increased does not count.
Market overview
💬 “What has risen the most over the last 4 hours?”
💬 “Where is funding negative right now?”
💬 “What is the maximum leverage on ETH for my account?”
Upscale has more than 150 markets: crypto, stocks, indices, Forex and commodities. For each one the agent sees the current price, the change over 1 and 4 hours, funding and the time of the next charge, maximum leverage by account phase, the fee, the spread and the market status — trading, paused or close-only. Funding is a periodic payment for holding an open position.
News and the market in one answer
ChatGPT and Claude can search the web, which complements Upscale data well:
💬 “What happened to Oracle today, and how has ORCL moved over the last 4 hours on Upscale?”
Web search is a feature of the agent app itself, not of Upscale, and its availability depends on the app and the plan. The agent does not receive candles or indicators from Upscale, so it cannot calculate RSI or similar indicators from platform data.
Instead of tallying stats in spreadsheets by hand, you ask a question about your trading and get an answer based on your own account's numbers.
Combined prompts: where the agent shines
The biggest impact comes from prompts where the agent combines account data, rules and markets, and then acts. All the examples below rely only on the data Upscale passes to the agent.
Morning briefing
💬 “Give me a report on all my accounts: room left before the limits, open positions and orders, funding on them and what has moved the most over the last 4 hours”
One prompt replaces going through several terminal tabs.
Position size within the limit
💬 “What size ETH position with a 3% stop can I open so that if the stop is hit I lose no more than half the room left before the daily limit? Show the calculation and wait for my confirmation”
The agent takes the start-of-day equity, the daily drawdown limit and the current price, calculates the size and places the order once you say “yes”. We covered how to choose your risk per trade in our guide to risk per trade in prop trading.
Pre-entry check
💬 “Before I short NAS100, check: is the market open or close-only, what leverage is available in my phase, what is the spread and when is the next funding charge”
A pause after a losing streak
💬 “I've closed three losing trades in a row. How much room is left before the daily limit, and what size is safe for me now?”
A prompt like this helps you stop and look at the numbers before trying to win it back.
Before the weekend
💬 “Show my open index positions and the markets that are currently close-only”
Positions in US and European indices and JP225 that are opened or increased in the last hours before the market closes on Friday are closed automatically. This prompt helps you decide in advance what to hold over the weekend. The rule is described in detail on the Challenge & Trading Rules page.
Cost audit
💬 “How much have I paid in fees and funding this phase? Which of my open positions are paying funding right now?”
Acting on several accounts
💬 “Close ETH positions on all accounts”
Picking markets at the weekend
💬 “It's Saturday: which markets are open right now, and where is funding negative?”
A trading journal in a spreadsheet
💬 “Export the closed positions history for the current phase to CSV: date, market, direction, result, holding time”
Claude Code and Cursor work with files on your computer, so the trading journal is built without manual copying. Bear in mind that the agent only sees trades from the account's current phase.
Learning on a demo account
💬 “On my demo account, use an open position to explain how leverage affects margin and what happens to equity if the price moves 2% against me”
For a beginner, the agent works as a tutor: it explains the mechanics using live account data rather than abstract examples.
Combined prompts save the most time: the agent gathers data from different sections and does the maths for you, while the decision stays yours.
How to write prompts so the agent gets it right
Upscale does not ask you to confirm each trade: an order sent by the agent goes for execution straight away. Whether the agent asks you before placing an order depends on the app itself, and when you connect it, you confirm that you take responsibility for its actions. So it is best to set the agent's rules in advance — in project instructions, personal settings or as the first message in the chat:
💬 “Rules for working with my Upscale accounts: before any order, show the account, market, direction, size, leverage, stop and take profit, and wait for my ‘yes’. Risk per trade — no more than 1% of equity. Always set a stop loss. After two losing trades in a day, do not suggest new entries”
This is not a hard block: a language model can make mistakes or misread a request, so check the parameters before confirming. A few rules for the prompts themselves:
- Name the account if you have several: the agent sees all accounts with the option at once.
- State levels as a number or a percentage of the current price. The agent does not receive a position's entry or liquidation price, so it cannot carry out “move the stop to breakeven”. Give a price instead: “move the stop to 80,000”.
- State size and leverage explicitly, not “all in” or “a bit more”.
- Try new scenarios on a demo account first.
A separate risk is prompt injection: hidden instructions in external data the agent reads, for example on a web page it opened while searching for news. In its connector help article, Anthropic advises connecting only trusted servers and watching for unexpected tool actions. Confirming each order protects you from this too.
Clear rules and specific prompts turn the agent into a careful executor that will not do anything extra without your consent.
Limitations
| The agent can | The agent cannot |
|---|---|
| See accounts, limits, positions, orders and the current phase's history | See accounts without the API Trading option or trades from previous phases |
| Place, modify and cancel orders, and close positions | Withdraw funds, buy challenges or options, or manage keys |
| Show price, 1- and 4-hour change, funding, leverage and spread for markets | Receive candles or calculate indicators such as RSI from Upscale data |
| Calculate your trading stats | Send price alerts |
| Work with all accounts that have the option at once | Work with only one selected account |
What else to keep in mind:
- The API Trading option's rules apply to the whole account, including manual trades: the 60-second profit adjustment and the 30% rule. The option can be added to accounts of $10,000 and above before the funded stage, and once enabled it cannot be removed.
- The agent is not for scalping. On top of the 60-second rule, there is a rate limit of 10 requests per second and 60 per minute.
- Access is granted to all accounts with the option at once. When connecting by server address you approve the permissions as a whole, and an API key always gives full access.
Knowing the limitations in advance, you give the agent tasks it can definitely handle and do not expect what it cannot do.
Key takeaways
An AI agent via MCP changes how you work with your account: instead of buttons and tabs, you write plain text, and ChatGPT, Claude, Claude Code or Cursor pull Upscale data, do the maths and place orders. Connecting takes a few minutes — by server address
https://mcp.upscale.tradeThe agent's main value lies in prompts that combine data. It calculates the room left before your challenge limits across all accounts, sizes a position to fit the daily limit, checks the market before entry, reviews your stats and closes everything with one command. You can copy the prompts from this article into the chat as they are and adapt them to your own markets and accounts.
The decisions stay with the trader. Upscale does not confirm each trade, so it is worth setting the agent's rules in advance: show the order parameters and wait for your “yes”. The API Trading option's rules apply to the account — the 60-second profit adjustment and the 30% rule — and new scenarios are best tested on a demo account first.
Start now: 👉 Upscale.trade | Telegram bot
Documentation: 📖 Upscale Docs
Follow us: 📺 YouTube | 𝕏 Twitter
Community: 💬 Telegram chat | 🎮 Discord
Frequently asked questions
Can I trade on Upscale through ChatGPT or Claude?
Yes. ChatGPT, Claude, Claude Code or Cursor can be connected via MCP at
https://mcp.upscale.tradeWhat is MCP in simple terms?
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external services. The AI understands your request in everyday language, and the Upscale MCP server turns it into actions on your account: fetching data, placing or cancelling an order.
Do I need to know how to code?
No. You write to the agent in plain text, for example: “How much room is left before the daily drawdown limit?” or “Buy $100 of BTC with 3x leverage, stop 5% below”. Code is only needed for API trading, where a bot runs without human input.
Is it safe to give an AI agent access to my account?
The agent places real orders, and Upscale does not ask you to confirm each trade. So give the agent a rule to show the order parameters and wait for your “yes”, check them before confirming and start on a demo account. You can disconnect the agent at any time on the MCP tab of the API and MCP settings.
Which accounts does the agent see?
Only accounts with the API Trading option — all of them at once; you cannot pick just one. The option can be added to accounts of $10,000 and above before the funded stage, and once enabled it cannot be removed. On a demo account it is enabled when the account is created.
How is MCP different from API trading?
With the API, a program trades by rules set in code, without human input. With MCP, you talk to an AI agent in plain text and it carries out your requests. Both require the API Trading option, and the same rules apply to the account: the 60-second profit adjustment and the 30% rule.
Can the agent draw a chart or send an alert?
No. The agent does not receive candles from Upscale, so it cannot calculate indicators such as RSI from platform data, and there are no price alerts. It does show the price change over 1 and 4 hours, funding, spread and maximum leverage for each of more than 150 markets.
Is the agent suitable for scalping?
No. Profit on a position closed less than 60 seconds after it was opened or increased does not count, and the request rate is limited to 10 per second and 60 per minute.
