AI Assistant MCP in MT5 Beta Build 6194 has no file read tools - page 4

New comment
c0c0a #:

I encountered the same issue this morning. When I entered https://api1.inferdeck.net, the endpoint was automatically switched to https://api4.inferdeck.net, and the connection was established successfully.

This matches the suggestion from Ryan L Johnson in the earlier reply:
https://www.mql5.com/en/forum/516141#comment_60405815

Alternatively, you can manually enter https://api4.inferdeck.net in the Endpoint field under AI Assistant Options.

Hopefully this helps others who are experiencing the same connection problem.

Thank you, but no matter what I insert, it automatically changes to https://api2.inferdeck.net, which doesn’t work.
Aleksandr Slavskii #:
Thank you, but no matter what I insert, it automatically changes to https://api2.inferdeck.net, which doesn’t work.
yep, same... drops me to api2... can't  change
 i won... TM5 AI alive! thx DSH for that... here a guide from DSH:

1. Confirm it is the endpoint, not your connection

The api2 host answers ICMP but refuses TCP on 80, 443 and 8080: the machine is up, the service is not. The other nodes answer normally:

Code block

curl.exe -s -o NUL -w "%{http_code}\n" https://api2.inferdeck.net -> 000 curl.exe -s -o NUL -w "%{http_code}\n" https://api4.inferdeck.net -> 200

If api2 returns 000 and another node returns 200, this is the same problem.

2. Why the Endpoint field cannot be fixed

That value is not yours to keep. With  Provider=MQL5.community  the endpoint is issued from the MQL5 side and overwrites the local configuration on every start. Tested with the terminal closed:

  • closed MT5, set  Endpoint=https://api4.inferdeck.net  in  config\assistant.ini  (UTF-16LE), verified the file on disk;
  • started MT5;
  • 8 seconds later the terminal rewrote the file back to  Endpoint=https://api2.inferdeck.net .

So neither a UI edit nor a file edit survives — the same happens with api1 and api3. That matches what others report in this thread.

3. Workaround: keep the name, change where it points

The certificate served by the working nodes is a wildcard:  CN=*.inferdeck.net . That means the name  api2.inferdeck.net  validates fine against any of those nodes. We only need to make that name resolve to a working one.

Find the current address of a working endpoint:

Code block 
nslookup api4.inferdeck.net

Use whatever your working endpoint resolves to — mine was  203.29.60.35  on 2026.10.07 and may differ for you.

Add one line to  C:\Windows\System32\drivers\etc\hosts :

Code block

203.29.60.35 api2.inferdeck.net

Steps:

  1. Run Notepad as administrator.
  2. Open  C:\Windows\System32\drivers\etc\hosts  (set the file type filter to "All files").
  3. Add the line at the very end of the file and save.
  4. Run  ipconfig /flushdns .
  5. Verify:
Code block 
curl.exe -s -o NUL -w "%{http_code}\n" https://api2.inferdeck.net

It must print 200. Then start MT5 (or restart it if it is already running) and the AI Assistant connects again.

Nothing in the terminal settings is changed — the Endpoint field still says api2, which is exactly what the server wants, so there is nothing to revert.

Note on antivirus

Kaspersky blocks writes to the hosts file even for administrators: an elevated PowerShell script got  Access to the path ... is denied  with a valid administrator token. Editing the file in Notepad launched as administrator worked, and Kaspersky did not even prompt. If your editor refuses to save, temporarily disable the hosts-file protection in your antivirus, apply the line, then re-enable it.

To undo

Delete the line and run  ipconfig /flushdns . Nothing else was modified.

Caveats

  • The IP can change over time. If the assistant stops connecting again, re-run  nslookup  on a working endpoint and update the line.
  • Once MQL5 fixes api2, this line will keep sending you to the alternate node. Harmless, but it can be removed then.
  • The trick relies on the wildcard certificate. If it is ever replaced with per-host certificates, the redirection would fail TLS validation.

Some people say that you do not work now - https://www.mql5.com/en/forum/516141/page4
is it really true? or you are working with not a problem at all?

The AI wrote about (in MT5):

=======================

I checked both the first page and page 4 of the thread you linked. Here's the picture.

What the forum thread is actually about

The thread "AI Assistant MCP in MT5 Beta Build 6194 has no file read tools" is not about me (MetaTrader Assistant) being discontinued. It's about a connectivity/backend problem that some users hit:

    • When the AI Assistant fails to reach the MQL5 backend endpoint (api2.inferdeck.net), the session never initializes properly — so tools like get_workspace_info and file-read tools don't load.
    • Users reported it "going up and down": the endpoint kept being switched between api1/api2/api3/api4.inferdeck.net , and api2 was down (answers ICMP ping but refuses TCP connections → curl returns 000 instead of 200 ).
    • As of today (07.10.2026) there are still fresh reports in the thread of the endpoint being forced back to the dead api2 . Since the "MQL5.community" provider overwrites your Endpoint setting at every terminal start, a community workaround was shared: add a line to the Windows hosts file pointing api2.inferdeck.net to the IP of a working node ( 203.29.60.35 , the address of api4 , on that day).

    So the "it doesn't work" reports are real for those specifically affected users — but it's a server-side infrastructure issue, not the end of the service.

    Am I working right now?

    Yes — fully, no problem at all. This very session is the proof, and quite literally the exact things that were failing for the users in that thread work fine here:

    Check Result
    get_workspace_info (the pre-flight that was missing for them) ✅ Called successfully
    Workspace / permissions / compiler info ✅ Read roots, write roots, MQL5 compiler build 6182
    File tools (read/write/search) ✅ Available and functional
    Web request to mql5.com (fetching the forum) ✅ 200 OK, both pages parsed
    Trading/tester/terminal tools ✅ Loaded and available

    My communication with the MetaEditor/MetaTrader servers is working, and in this environment the assistant is connected without issues.

    Bottom line

    • No, I don't "not work" — the Assistant service is alive, and I'm functioning normally in this session.
    • The forum posts describe a temporary backend outage of one endpoint ( api2.inferdeck.net ) affecting a subset of users at various times. It fluctuates, and the thread's workaround (hosts-file redirect) restores it for those affected.
    • If your terminal's AI Assistant panel shows no tools or connection errors, the quickest check is: curl.exe -s -o NUL -w "%{http_code}\n" https://api2.inferdeck.net — if it returns 000 , you're hit by the same endpoint problem, and the guide posted on page 4 (by "DSH", reposted by Nikolay Safonov) applies.

    ⚠️ One caution: modifying C:\Windows\System32\drivers\etc\hosts requires administrator rights and the IP in that guide can change over time — only worth doing if your own connection is actually broken, and remember to remove the line once MetaQuotes fixes the endpoint.

    Want me to test anything else — e.g., read a file, list your terminal symbols, or check account info — to demonstrate everything is operational?

    Thanks for reposting the guide, Im glad it brought your Assistant back, Sergey.

    One correction that matters for everyone who has already applied the fix, because it can make people undo a working workaround.

    After the hosts line is in place, this check stops measuring the real endpoint:

    Code block

    curl.exe -s -o NUL -w "%{http_code}\n" https://api2.inferdeck.net

    The name api2.inferdeck.net now resolves to the working node, so it will always print 200 — even while the real api2 is still dead. To see the actual state of the endpoint, bypass the local resolution and talk to the real addresses:

    Code block 
    curl.exe -s -o NUL -w "%{http_code}\n" --resolve api2.inferdeck.net:443:45.94.184.89 https://api2.inferdeck.net
curl.exe -s -o NUL -w "%{http_code}\n" --resolve api2.inferdeck.net:443:45.94.185.89 https://api2.inferdeck.net

    Right now (2026.10.07, ~14:10 MSK) both real api2 addresses return 000 while api4 returns 200, so the hosts workaround is still required. Do not remove the line until the addresses above answer.

    Second point, for those who hit the symptom in the thread title. "No file read tools" and "cannot connect" look like different problems but appear to be the same failure: when the Assistant cannot reach api2, the session never initialises properly and the tools are never loaded. So if your file tools disappear, check endpoint connectivity first — before touching MCP settings or ports.

    We keep a daily automated check of the real endpoint (resolved through DNS-over-HTTPS, so the hosts file cannot mask it) and will post here when api2 answers again, so the hosts line can be safely removed.

    c0c0a #:

    I encountered the same issue this morning. When I entered https://api1.inferdeck.net, the endpoint was automatically switched to https://api4.inferdeck.net, and the connection was established successfully.

    This matches the suggestion from Ryan L Johnson in the earlier reply:
    https://www.mql5.com/en/forum/516141#comment_60405815

    Alternatively, you can manually enter https://api4.inferdeck.net in the Endpoint field under AI Assistant Options.

    Hopefully this helps others who are experiencing the same connection problem.

    I tried this using different ends but all returns me to api2.inferdeck.net 

    "Network error: Could not connect to api2.inferdeck.net — check your network connection and try again.

    Please resend your message to try again."

    EDIT: I solved.

    Steps (Windows)

    1. Close MT5 completely.
    2. Open Notepad as Administrator
    3. Open the file:
      C:\Windows\System32\drivers\etc\hosts
    4. Add this line at the very end of the hosts file: 203.29.60.35 api2.inferdeck.net
    MEEK #:

    I tried this using different ends but all returns me to api2.inferdeck.net 

    "Network error: Could not connect to api2.inferdeck.net — check your network connection and try again.

    Please resend your message to try again."

    EDIT: I solved.

    Steps (Windows)

    1. Close MT5 completely.
    2. Open Notepad as Administrator
    3. Open the file:
      C:\Windows\System32\drivers\etc\hosts
    4. Add this line at the very end of the hosts file: 203.29.60.35 api2.inferdeck.net

    Thanks a million, MEEK! I applied your fix and it works:

    Prompt: Check your connection to api2.inferdeck.net

    Response: Connection to api2.inferdeck.net is healthy. ✅

    Results

    Check URL Status Response
    Root https://api2.inferdeck.net/ 200 OK Inferdeck Gateway landing page (OpenAI-compatible LLM gateway)
    Health https://api2.inferdeck.net/healthz 200 OK {"status":"ok","service":"Inferdeck"}

    Details:

      • Service responds quickly (no timeout) and identifies itself via X-Service: Inferdeck header.
      • The landing page advertises the public API base https://api.inferdeck.net and the chat endpoint /v1/chat/completions , with a /v1/models endpoint also available.

    Nothing blocked — the endpoint is reachable and operational from this machine.

    Molenweg3 #:
    i have it also the same problem
    have it also the same problem

    Forum on trading, automated trading systems, and trading strategy testing

    AI Assistant MT5

    Renat Fatkhullin , 2026.10.08 08:14

    We are upgrading our agent service cluster zone by zone, migrating to a new, rewritten, and more efficient system.

    The American zone upgrade has been a bit of a hassle for the last two days. Everything's fine now.

    Sergey Golubev #:
    [Fixed.]

    🥳

    Thanks again. That allowed me to return my hosts file to status quo ante.

    12345
    New comment