"integrate the whole thing with VS Code"
does this mean you also edit your AFL in VSC? Setting it up this way means that the LLM can also do the editing for you?
So Goose Desktop allows one to switch between LLM's easily? That would be pretty useful.
I am just starting with this. The LLM I use has its own Desktop application. I just use the Amibroker editor. Only work I need to do is to apply the changes to the AFL that the LLM suggests
AFL editor is good, and if you have the ultimate pack and are using it because AB has some optimize prompt but for a noob, there rest of the things are very easy with VSC compared to anything else.
So "continue" extension is nice, you can add MCP server as well as the LLM endpoints to it. There are many different extensions, depending on what you want to do with GUI options.
If you are doing other python things, then all this can be taken care here. You wont need many different editors. I have yet to try AFL, but the built-in one is nice so never tried but i am sure you it wont be so bad. others on the forum have used external editors.
thanks, I might implement some of your suggestions later although my simple setup is basically good enough.
I am glad for these LLM's because it helped me understand your answers 
Hello Tomasz, I attempted to configure LLM environment in 7.20 and received an error following TestLM
Am I doing something wrong or is Gemini-2.5-flush no longer available?
Also, Do I need AFL code wizard license to use AI assistant features?
The documentation says that it is required. I am happy to buy the license if you confirm.
Kind Regards
Richard
This is GOOGLE error, not AMiBroker.
They simply retire old model(s).
Do as the message says, instead of gemini-2.5-flash, type gemini-3.6-flash into "Model" field.
Hello sandeep,
first, study ...
https://gitlab.com/amibroker/mcpserver
For my understanding MCP Server is the interface between Amibroker and LLM. The LLM acts as MCP Client for communication to Amibroker via MCP Server process. I did some modifications and try to document my setup here.
Download python 3.11.9 - April 2, 2024
https://www.python.org/downloads/windows/
Install python
Open/Start python-3.11.9-amd64.exe
...Customize installation
My python directory
C:\Python311
It is absolutely essential to check Windows path env to locate python, pip etc. from any stage. If path is not expanded during installation you have manually add infos via Windows tools.
Start Cmd and check
path (return)
...
Check python is running
Start Cmd and check
python --version (return)
Install python dependencies through Cmd
pip install fastmcp pywin32 psutil requests html2text starlette uvicorn (return), ... wait some time ...
Create directories
C:\mcp
C:\mcp\mcpserver
Place mcp_my_server_http.py into C:\mcp\mcpserver
Moderator comment: the steps listed below ARE NOT needed
C:\mcp\mcpdata
My mcp_my_server_http.py script is located in C:\mcp\mcpserver
Modify mcp_my_server_http.py and adjust path
...
#destination path for various operations
#dest_path = os.getcwd() + "\mcp_data\"
dest_path = "C:\mcp\mcpdata\"
...
Open Cmd “As Administrator”, so color coding is missing …
Keep MCP Server running, only terminate after finishing LLM env
In the command prompt, type:
cd C:\mcp\mcpserver
python mcp_my_server_http.py
Without Admin Context, ... does not work in my env
Through Admin Context MCP Server is working
Check MCP Server is up and running in Cmd
netstat –a
At this point any AI Shell, like LM Studio, on-premise/local installed should be able to connect the MCP Server using implemented Amibroker interface/tools.
Best regards,
Peter
Thank you very much for your support. It is the reason I value this community so highly. I would also like to extend my sincere gratitude to Tomasz for developing such outstanding software and maintaining this invaluable support forum.
These are quite good instructions, some steps are not needed.
C:\mcp\mcpdata folder.mcp_my_server_http.py and you do NOT need to modify paths.Python script "just works" out of the box without any modifications. This is so because Python script is written so it creates data subfolder mcp_data INSIDE current working directory by itself and automatically sets the path to the working directory + \mcp_data
Therefore these steps can be just skipped. Instead of modyfing paths and files,
you should change current working directory to MCP server dir and then launch it in proper directory:
cd C:\mcp\mcpserver
python mcp_my_server_http.py
This way is way better as any files/dirs are NOT written to "system" directory that would happen if you start python inside "C:\Windows\System32" as shown above.
Also this Python script is not necesarily tied to Python 3.11. Personally I am running 3.13 and it works and in fact it should with any other Python version higher than 3.11 (which is already 2 years old)
Hello Tomasz,
I am sorry for asking this basic question but, what Gemini tool do I need to run to "connect/talk to the AB session?
So far, I installed Anaconda3 Python and mcp server from Github.
When I prompt Gemini in Firefox web browser, it seems like it is not talking to AB.
Kind Regards
Richard Pacholski
Hi Tomasz,
I just asked Gemini what is this tool and it said - Bionic LM Studio.
I have it installed but, I thought that is only needed for local LM models.
Do I need to configure Bionic to talk to Amibroker MCP server?
Kind Regards
Richard Pacholski
No, it won't work just from the Gemini web page because it is not MCP client. You need a tool that is MCP client (the program that talks locally to MCP server and exposes tools to LLM).
For the video that I have shown earlier I have used Goose: https://goose-docs.ai/
You can configure it to work with both cloud and local LLMs and custom MCP servers.
Please note that these features are for advanced users, who already use agents and want to expand, because running agents bring certain risks, if you enable LLM to run tools on your local machine it can be potentially security risk, as it can access your private stuff on your computer. So I would advice against using it unless you run on separate machine without any of your private data.
Hello richpach,
LM Studio has two different deployments. Bionic Version does work with MCP in different way, you have to setup "Skills", link your MCP Server...
So for my env it works download/install, you configure MCP Client direct in LM Studio (NOT Bionic):
LM-Studio-0.4.21-2-x64.exe
best regards,
Peter
according to the LLM's, LM studio requires you to download the models on your computer. What I was told is that if you want to switch between LLM's (meaning a third-party chat app that allows you to switch between LLM's) in the cloud you need to use other tools.
As I wrote before, to use cloud LLMs, you need MCP client, like Goose mentioned earlier.
MCP client (aka "harness") interfaces between LLM and MCP server.
my response was actually about LM studio because @nsm51 and @Pietro were using that. I was in the process of installing LM studio but then it asked about which model to download (if I remember correctly).
I thought LM studio would act as a MCP host and would allow me to switch between Anthropic, Google, xAI etc. MCP clients in the cloud but it seems there are better tools for that. I will just use 1 LLM for now. That has its own desktop interface (MCP host) and I'm working in the cloud. Not downloading models on my desktop.
So basically I thought LM studio would act as the MCP host that could contain multiple MCP clients. But as I understand (told by AI) there are better tools for that. I was suggested to tools like LibreChat. That is what I am told by AI so possibly @nsm51 and @Pietro know more about that.
For now I will just use one AI company that has its own MCP host + client
Hello Tomasz,
Thanks you for your reply and for pointing out risks involved with using AI agents. I realize that, I am not an advanced user, even though I have been using Amibroker for many years. I will not mention the number because this my reveal how old I am. I enjoy learning new skills and I am very happy that following your advise I got it working with "goose". Now, I understand a little bit better how MCP concept works with Amibroker. Perhaps to avoid any risks in the future, I will use it with local LLM model. Do you have any recommendation which one is the best and what size of model is sufficient for Amibroker AI functionality?
Kind Regards
Richard Pacholski
Hello Peter,
Thank you for your advise. I struggled a bit with LM Studio. I downloaded the same release as on your screen capture but I could not see all the menu option (on the right hand side) to install Integrations. Eventually, I got it working with "goose" as per Tomasz's suggestion.
Thanks again for your help.
Kind Regards
Richard Pacholski
Hello Richard,
AI infrastructure with LM Studio is on-premise/local setup without Cloud AI services. Any "Chat" in LM Studio is linked to a Model you import first in LM Studio and link with your corresponding "Chat".
{
"mcpServers": {
"AB-MCP-Server": {
"url": "http://127.0.0.1:8000/mcp"
}
}
}
overwrite current mcp.json, save this config file. This MCP Client talks to MCP Server, Tomasz developed and has to be up and running. The Tools are imported from the MCP Server is publishing.
Gemma 4 family of models (12B or larger) and Qwen 3.5-3.8 (27B) are worth trying.
