ELDRA × AI Agents for Beginners: From Chatting to Setup, Safety Guards, and Backtesting

📢 Author / advertising disclosure: ELDRA is an EA built by this site's owner (Haruki Hosono), and the MCP used in this article is also provided by this site. This is not a neutral, third-party article. The Fintokei, The5ers, and Funded7 links in the body are affiliate links, and buying through them gets you unlimited ELDRA use on that account.

ELDRA has 438 configurable parameters. You don't need to memorize all of them to run it, but plenty of people get stuck simply not knowing where to start.

So we built a system that gives an AI agent access to ELDRA's parameter dictionary and diagnostic tools. This article walks through the steps in order, aimed at people who aren't especially technical, from the first connection all the way up to "right before you run it on a prop firm account."

The overall flow

At every step, all you do is "ask in plain language" and "check the results." The AI handles the detailed operations for you.

Getting the terminology straight first

Step 1: Set up an AI agent

Claude Code is recommended. The reason is that it can run commands on your computer, not just chat. That's what lets you hand off Step 7 — "run the backtest, read the results, adjust the settings" — to the AI.

• You'll need a paid Claude plan (Pro or above), or an API account
• Follow the installation instructions on Anthropic's official Claude Code page (on Windows, it's just one line pasted into PowerShell)
• Once installed, type claude in PowerShell to launch it

💡 A chat-only AI (a chat app that supports MCP) can still get you through Steps 4-6 — "discuss it and build a .set file." Only the automation in Step 7 requires an agent that can run commands.

Step 2: Connect prop-memo's MCP

Once Claude Code is running, just send this exact line:

Read https://prop-memo.com/api-docs and connect prop-memo's MCP

The AI will read the connection instructions page and configure it itself. If you're doing it manually, this is the equivalent PowerShell command:

claude mcp add --transport http propmemo https://prop-memo.com/mcp

If it says "please restart," just close Claude Code and launch it again with claude.

ELDRA's tools require no sign-up and no API key. Only if you want to hand off your own challenge and payout records to the AI as well do you need to add an API key later (see "API Key" on the connection page).

To check that the connection worked, try asking:

What kind of EA is ELDRA? Who is it a good fit for, and who isn't it for?

If you get back an overview of ELDRA along with the fact that the author is this site's owner, the connection is working.

Step 3: Get ELDRA and install it on MT5

Backtesting and demo accounts are free, unlimited use (live accounts get a 3-day trial).

1. Download ELDRAf.ex5 (or .ex4 for MT4) from the ELDRA overview page
2. In MT5, go to File → Open Data Folder
3. Put ELDRAf.ex5 into MQL5\Experts inside the folder that opens
4. Restart MT5, and ELDRAf will show up under "Expert Advisors" in the Navigator

⚠️ Prop firm accounts are not covered by the free tier, even demo accounts. To run it on a prop firm account, you need to authenticate via one of: purchasing through an affiliate link (Fintokei, The5ers, Funded7), the community, or a one-time purchase on Gogojungle. Backtesting is free in every case, so try working through Step 7 before you buy anything.

Step 4: Ask the AI whether your strategy can be built

ELDRA isn't "an EA that wins automatically once you install it." You decide where to enter; the EA handles exits, lot sizing, and stopping before you hit the disqualification line.

So the first thing to do is put your strategy into words and ask the AI.

Can ELDRA build a rule like "buy when RSI drops below 30 and then comes back up"?
I want a 20-pip stop loss and a 40-pip take profit.

The AI searches ELDRA's parameter dictionary for the relevant settings. For this example, you'd get an answer like this:

ELDRA's parameters have numbers like "50-2" attached, and the MT5 settings screen lists them with the same numbers. You can also ask the AI "what is 50-2?" using just the number.

💡 The AI is set up not to claim a feature exists if it's not in the dictionary. If it says "that condition doesn't exist in ELDRA, but here's the closest thing," that's an honest answer. Entering off a signal from an indicator you already have (37. Indicator Entry Settings) can often build the same thing, so it's worth asking about that too.

Step 5: Add safety guards matched to your prop firm's rules

The single most wasteful way to fail on a prop firm account, aside from whether your strategy is any good, is getting disqualified by hitting the max-loss or daily-loss rule. This is the part it makes the most sense to hand off to the AI.

Give me a recommended ELDRA setup for Fintokei's ¥20M account (Sapphire).

The AI pulls that plan's max loss (10%) and daily loss (5%) from this site's pricing database and returns a setup like this:

You'll also get that firm's specific caveats along with it. For Fintokei, that means things like "total risk across all simultaneously open positions is capped at 3% of the initial balance" and "using someone else's EA without adjusting it is prohibited."

⚠️ This is not "a setup that wins." It's "a setup designed to avoid rule violations and disqualification." At firms where the max-loss line moves upward over time (trailing type), you'll need to update the floor amount yourself as it moves.

Step 6: Build a .set file and load it into MT5

Once you have Step 4 (the strategy) and Step 5 (the safety guards) sorted out, have the AI combine them into one file.

Combine the RSI condition from before with the Fintokei safety guards
into a .set file, and save it to the desktop as fintokei_rsi.set

How to load it into MT5:

1. Drag ELDRAf onto a chart to open the settings screen
2. Click "Load" in the bottom right of the "Inputs" tab
3. Select the saved .set file and click OK

If you already have a .set file you built yourself, hand it to the AI and ask it to "check this setting for mistakes." It will flag common errors — like "no exit method at all," "a condition set to ON that has no effect," or "martingale with no stop loss" — along with a suggested fix, and hand back a corrected .set file too.

Step 7: Verify with a backtest

Check how the setup you built performs against historical market data. There are two ways to do this.

A. Run it yourself (works with any AI)

In MT5, go to View → Strategy Tester, choose ELDRAf as the expert, pick the instrument, date range, and timeframe, load the .set file under the "Inputs" tab, and click "Start." That's it.

B. Have the AI run it (Claude Code)

Run a backtest on XAUUSD M15 from January to December 2025 using
the .set file from before. Tell me the net profit, max drawdown,
and number of trades.

Claude Code handles the following automatically:

1. Builds an ini file for the backtest
2. Launches MT5 with that ini file — MT5 closes automatically once the test finishes
3. Reads the resulting report and summarizes it in plain language
4. If you ask "widen the stop loss a bit," it adjusts the settings and goes back to step 1

It will ask "is it OK to close MT5?" before running. This is because the tester won't run while MT5 is already open. If you have an MT5 instance actively trading, it's safer to do this on a separate installation (MT5 installed in a different folder).

The most common snag: instrument names

Even for the same gold, the instrument name differs by broker.

If you specify an instrument name that doesn't exist on the account MT5 is currently logged into, the tester ends without doing anything, and no report is produced. The AI looks for the actual instrument name from MT5's folder, but if it can't find one, it will ask you to "tell me the instrument name shown in Market Watch." When that happens, just give it the exact name shown in MT5's Market Watch window.

One more thing: the tester sometimes won't launch on an MT5 instance that has never logged into the account. It's safer to open MT5, log into the account, and open a chart once before closing it, the first time.

Step 8: Optimize by "deciding on the first half, verifying on the second half"

You can also ask the AI to run optimization, automatically sweeping something like the stop-loss distance to find the best-performing value.

Optimize the stop loss between 10 and 40 pips in 5-pip increments

But this is where beginners most commonly get tripped up. The value that fit the past best usually misses the future (overfitting). Always follow this procedure:

1. Split the period into a first half and a second half (e.g. 2023-2024 and 2025)
2. Optimize and pick a value using only the first half
3. Keep that value fixed and backtest it on the second half, to check whether performance holds up

Just ask the AI directly: "optimize on 2023-2024 and verify on 2025," and it will run this exact procedure. Discard any value that falls apart in the second half.

Run the final check with the "Every tick" model, which has the highest tick accuracy (just ask "run the final check with every tick").

Step 9: Demo → live

1. Once you're satisfied with the backtest, run it on a demo account for a few days first and check that entries and exits are happening as expected
2. Before putting it on a live (prop firm) account, run the final .set file through "check this for mistakes" one more time
3. Keep the lot size the same during evaluation and after funding (some firms treat trading aggressively during evaluation and then scaling back after funding as grounds for a rules violation or extra payout scrutiny)

Things not to do

Example prompts you can use as-is

Can ELDRA build "buy when price breaks above the London session high"?

Give me a safety setup for The5ers' High Stakes $100K. Don't use averaging down.

Check this .set (I'll hand you the file) for whether it's set up in a way
that's likely to hit the daily loss limit.

Optimize on USDJPY H1 for 2022-2024, verify on 2025.
Try take-profit candidates from 20 to 60 pips in steps of 10. Put the results in a table.

Take the current setup, change just the take-profit to 30 pips, and save it as a new .set file.

FAQ

Q. Do I need programming knowledge?

No. All you do is ask in plain language and check the results the AI produces (tables of settings, backtest numbers). The AI handles the commands and file operations.

Q. Does this cost money?

ELDRA's backtesting and demo accounts are free, and prop-memo's MCP is also free with no sign-up required. Using an AI agent (Claude Code) requires either a paid Claude plan or API usage fees. Running ELDRA on a prop firm account requires authentication via an affiliate-link purchase, the community, or a one-time Gogojungle purchase.

Q. Should I just trust the settings the AI builds?

Don't. The AI builds settings based on ELDRA's dictionary and prop firm rules, but it can't judge whether the strategy itself is any good. Always verify with your own eyes via backtesting and a demo account before using it.

Q. Does this work with MT4 too?

The parts where you discuss and build a .set file (Steps 4-6) work the same way with the MT4 version. The feature that has the AI run backtests automatically (Step 7B) is for MT5.

Q. I'm worried the AI will close my MT5 while I'm trading

The AI checks whether it's OK to close MT5 before running a backtest. If you have an MT5 instance actively trading, it's safer to install a separate copy of MT5 dedicated to backtesting.

Q. What if asking the AI doesn't solve my problem?

There's a community of ELDRA users, including a LINE open chat and the CAMPFIRE community "Hosono P's FX Base." Links to both are on the ELDRA overview page.

Summary

• Connecting an AI agent (Claude Code) to prop-memo's MCP gives you access to ELDRA's dictionary of 438 parameters, setting diagnostics, prop-firm-matched safety guards, and automated backtesting
• The flow is: connect → install ELDRA → ask whether your strategy can be built → safety guards → .set file → backtest → optimize by checking first half / second half → demo → live
• The AI can build "a setup that correctly implements your strategy" and "a setup that's less likely to get disqualified." Whether it actually wins comes down to the strategy and the backtest
• Backtesting is free, so you can go this far before ever buying a challenge

For detailed connection steps, see prop-memo API / MCP. For ELDRA's track record and how to get it, see the ELDRA overview page. For EA rules by prop firm, see Prop firm EA rules compared.