MetaEditor, Open AI and ChatGPT - page 10
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
A local neural model can be run as a chat room or as a server. You connect to the server via API and JSON.
I recommend ollama or llama.cpp as the fastest and easiest (in c++, not python), and the most customised for GGUF.
LMStudio is the most convenient interface for chat.
Is it possible to have an MCP official please?
Is it possible to get an official MCP document?
For an MCP developer?
https://modelcontextprotocol.io/specification/2025-11-25
https://github.com/modelcontextprotocol/modelcontextprotocol
https://www.anthropic.com/engineering/code-execution-with-mcp
For connecting ready-made MCPs I read https://qwenlm.github.io/qwen-code-docs/en/users/features/mcp/
In trading, I believe that AI won't be able to help you if you don't understand anything about trading. In other words, the AI will do what you tell it to do, so you must be familiar with all aspects to give it correct orders and benefit from it. I've been trading since 2013, and if I didn't understand the programming languages of expert advisors and indicators, AI wouldn't have been of any use to me at all. However, it does help me by correcting some spelling mistakes in my code.
How to connect AI agents to MetaTrader 5 via MCP
How to connect AI-agents to MetaTrader 5 via MCP
It is important to clarify one point for this forum
Algotrading is one of the few areas where AI gives surprisingly little advantage right now, although it would seem - where else but here, pure maths, pure data.
The reason is the very nature of the market. All public data on which LLMs are trained, by definition, contains knowledge that the market has already arbitraged. If a strategy is well described in a book, article or forum - it either doesn't work anymore, or it never did. Really working approaches don't make it into the training corpus for a very simple reason - nobody publishes them. It's the survivor effect in reverse: only things that don't offer an advantage remain in the public domain. It turns out that the more a model has read about trading, the more confidently it reproduces already dead patterns.
Next is non-stationarity. The model is trained on a distribution that has already changed by the time it is used. The market is not a game of Go with fixed rules. It is a game where the rules change every time someone finds a new pattern and everyone else starts reacting to it. Any inefficiency discovered is a self-defeating prophecy.
And most profound of all is the adversarial nature of the environment. Unlike picture classification or text generation, where the task is benevolent, the market actively resists being recognised. Every signal you make is someone else's direct loss, and that other party is not sitting idly by. If a pattern becomes obvious to a model trained on public data, that means it's obvious to everyone, and that means it's gone by the time you hit the button.
That's why AI is useful in algo-trading right now, not at all where they are trying to apply it. It is bad at generating alpha, but it is great at making infrastructure: backtest engines, visualisers, parameter optimisers, news parsers, and pipelines for hypothesis testing. The bottle neck is shifting in the process. It used to be implementation - now it is idea and intuition. And an idea in trading is something that cannot be learnt from a case. You need to see it yourself, preferably to see what others do not see yet.
When (and if) AI gets to real trading systems, it will not happen through the prompt "write me a profitable strategy in MQL5". It will be a specialised model trained on market data with the right RL lens by some Renaissance or Citadel. And on the same day, when it works, it will be locked in a safe, not posted in the API. So the forumers waiting for AI to "write the grail" will have to wait a long time, and not where they are looking.
my advice to those who are confused: don't try to catch up with the narrow specialists in their verticals. That race is already lost. В
Extremely doubt the above.
Creating a ToR for AI without understanding what the AI has returned to you is a path to nowhere.
The same question, what should be the size of the promt that the AI would generate a medium-sized project.
Well, you have written a 100500-page promt task, the AI has given you the code, there are errors, what next?
The example above describes Shovels and Excavators.
Working as a master with a shovel you will make the hole that you need size, accuracy of edges, depth, and at the first time and without damaging communications.
Working as an excavator you can not achieve this.
In general, this comparison in my opinion is not the most successful....
AI copes well with data analysis, but to create something, especially from programming, I think it will not be soon.
In any case, when the code created by AI brings 100500 losses, the customer will cross himself 100500 times and hire a person capable of fixing errors.
And as practice shows, picking someone else's code, especially AI, is much more expensive than writing code from scratch.
There have always been amateur and professional programmers, so AI is a light amateur. When you need something working and reliable, it is the way only to the pros....