Mintplex-Labs/anything-llm
1
0
Fork 0
mirror of https://github.com/Mintplex-Labs/anything-llm.git synced 2026-07-19 14:13:35 -04:00
Code Issues 328 Packages Projects Releases 10 Wiki Activity

[GH-ISSUE #3719] [Question]: MCP docker commands on failing to start on Windows? #2396

New Issue
Closed
opened 2026-02-22 18:29:29 -05:00 by yindo · 3 comments
yindo commented 2026-02-22 18:29:29 -05:00
Owner
Copy Link
Copy Source

Originally created by @kfsone on GitHub (Apr 25, 2025).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/3719

Originally assigned to: @timothycarambat on GitHub.

How are you running AnythingLLM?

AnythingLLM desktop app

What happened?

  • Docker 4.40,
  • Windows 11,
  • Anything LLM 1.8.0-r2,

I fetched several docker mcp images (https://hub.docker.com/catalogs/mcp), tested them all from Python and found them to be working, integrated them all to Roo Code without hassle.

I was only able to get two to work with Anything LLM, and then not at the same time. (mcp/fetch, mcp/git)

Others just don't work, e.g. the mcp/duckduckgo instance - although it works in Roo/Cursor/Claude/from Python.

This config:

Win|PS> more .\anythingllm_mcp_servers.json
{
  "mcpServers": {
    "fetch-web-page": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "--init",
        "-e",
        "DOCKER_CONTAINER=true",
        "mcp/fetch"
      ]
    }
  }
}

Clicking refresh does not detect a new MCP server. Restarting AnythingLLM detects it but reports an error, that the connection closed. If I add

     "anythingllm": {
        "autoStart": true
      }

to the config and refresh, the server disappears from the list. If I restart AnythingLLM twice, on the second start the server finally shows as "On".

Image

If I remove the autoStart and refresh:

Image

But now if I restart it, again, we're back to "on".

Are there known steps to reproduce?

  • install docker
  • install anything llm
  • restart
  • launch docker desktop,
  • cmdline: docker pull mcp/duckduckgo
  • cmdline: docker pull mcp/fetch
  • exit anythingllm
  • edit anythingllm mcp json
{
  "mcpServers": {
    "search-the-web": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "--init",
        "-e",
        "DOCKER_CONTAINER=true",
        "mcp/duckduckgo"
      ]
    }
  }
}
  • restart anything llm
  • go to mcp servers list,
    -> no mcp servers found,
  • exit anythingllm
  • edit the above config and replace mcp/duckduckgo with mcp/fetch
  • start anythingllm
  • go to mcp servers list
    -> one mcp server found -- if not, exit and restart anythingllm once,
    ^- stop if no server shows
  • exit anythingllm
  • edit the mcp json to contain two servers:
{
  "mcpServers": {
    "fetch-a-page": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "--init",
        "-e",
        "DOCKER_CONTAINER=true",
        "mcp/fetch"
      ]
    },
    "search-the-web": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "--init",
        "-e",
        "DOCKER_CONTAINER=true",
        "mcp/duckduckgo"
      ]
    }
  }
}
  • restart anythingllm
    -> it's ok to cry a little

NOTE: If you remove the remove ("--rm") and repeat this, you will see that it starts both containers, which makes me think it's getting confused by them having the same "command".

Originally created by @kfsone on GitHub (Apr 25, 2025). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/3719 Originally assigned to: @timothycarambat on GitHub. ### How are you running AnythingLLM? AnythingLLM desktop app ### What happened? - Docker 4.40, - Windows 11, - Anything LLM 1.8.0-r2, I fetched several docker mcp images (https://hub.docker.com/catalogs/mcp), tested them all from Python and found them to be working, integrated them all to Roo Code without hassle. I was only able to get two to work with Anything LLM, and then not at the same time. (mcp/fetch, mcp/git) Others just don't work, e.g. the mcp/duckduckgo instance - although it works in Roo/Cursor/Claude/from Python. This config: ``` Win|PS> more .\anythingllm_mcp_servers.json { "mcpServers": { "fetch-web-page": { "command": "docker", "args": [ "run", "-i", "--rm", "--init", "-e", "DOCKER_CONTAINER=true", "mcp/fetch" ] } } } ``` Clicking refresh does not detect a new MCP server. Restarting AnythingLLM detects it but reports an error, that the connection closed. If I add ``` "anythingllm": { "autoStart": true } ``` to the config and refresh, the server disappears from the list. If I restart AnythingLLM *twice*, on the second start the server finally shows as "On". ![Image](https://github.com/user-attachments/assets/dab77fb6-cbe6-4fb2-8108-7799e5f82c1f) If I remove the autoStart and refresh: ![Image](https://github.com/user-attachments/assets/7c881ea6-16ed-4036-a717-394022346f62) But now if I restart it, again, we're back to "on". ### Are there known steps to reproduce? - install docker - install anything llm - restart - launch docker desktop, - cmdline: `docker pull mcp/duckduckgo` - cmdline: `docker pull mcp/fetch` - exit anythingllm - edit anythingllm mcp json ``` { "mcpServers": { "search-the-web": { "command": "docker", "args": [ "run", "-i", "--rm", "--init", "-e", "DOCKER_CONTAINER=true", "mcp/duckduckgo" ] } } } ``` - restart anything llm - go to mcp servers list, -> no mcp servers found, - exit anythingllm - edit the above config and replace `mcp/duckduckgo` with `mcp/fetch` - start anythingllm - go to mcp servers list -> one mcp server found -- if not, exit and restart anythingllm once, ^- stop if no server shows - exit anythingllm - edit the mcp json to contain two servers: ``` { "mcpServers": { "fetch-a-page": { "command": "docker", "args": [ "run", "-i", "--rm", "--init", "-e", "DOCKER_CONTAINER=true", "mcp/fetch" ] }, "search-the-web": { "command": "docker", "args": [ "run", "-i", "--rm", "--init", "-e", "DOCKER_CONTAINER=true", "mcp/duckduckgo" ] } } } ``` - restart anythingllm -> it's ok to cry a little NOTE: If you remove the remove ("--rm") and repeat this, you will see that it starts both containers, which makes me think it's getting confused by them having the same "command".
yindo added the bugpossible bugDesktopinvestigating labels 2026-02-22 18:29:29 -05:00
yindo closed this issue 2026-02-22 18:29:29 -05:00
yindo commented 2026-02-22 18:29:29 -05:00
Author
Owner
Copy Link
Copy Source

@timothycarambat commented on GitHub (Apr 25, 2025):

So this config

{
  "mcpServers": {
    "fetch-a-page": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "mcp/fetch"
      ]
    }
  }
}

No issues, but also totally standalone the mcp/duckduckgo server never starts - it pulls and tracks in docker, but AnythingLLM can never connect to it properly

Its worth noting that with docker even though the MCP SDK should spawn a new process for each, so stdio communication works, that could be an issue with multiple -i servers?

Outside of that, there is also a bug with bulk-pruning of servers on Windows, which will require a patch and will be out next week.

I can get fetch to work fine, as well as fetch + slack and some others. Duckduckgo is the one, that when enabled, causes issues.

Worth noting that when using the same servers via uvx command - I have no issues.

@timothycarambat commented on GitHub (Apr 25, 2025): So this config ```json { "mcpServers": { "fetch-a-page": { "command": "docker", "args": [ "run", "-i", "--rm", "mcp/fetch" ] } } } ``` No issues, but also totally standalone the `mcp/duckduckgo` server never starts - it pulls and tracks in docker, but AnythingLLM can never connect to it properly Its worth noting that with `docker` even though the MCP SDK should `spawn` a new process for each, so stdio communication works, that could be an issue with multiple `-i` servers? Outside of that, there is also a bug with bulk-pruning of servers on Windows, which will require a patch and will be out next week. I can get fetch to work fine, as well as fetch + slack and some others. Duckduckgo is the one, that when enabled, causes issues. Worth noting that when using the same servers via `uvx` command - I have no issues.
yindo commented 2026-02-22 18:29:30 -05:00
Author
Owner
Copy Link
Copy Source

@timothycarambat commented on GitHub (Apr 28, 2025):

Fixed the crash - patch will be live in 1.8.1. Continuing to investigate this specific server config stopping others from booting - which seems to be MCP related at the current moment

@timothycarambat commented on GitHub (Apr 28, 2025): Fixed the crash - patch will be live in 1.8.1. Continuing to investigate this specific server config stopping others from booting - which seems to be MCP related at the current moment
yindo commented 2026-02-22 18:29:30 -05:00
Author
Owner
Copy Link
Copy Source

@timothycarambat commented on GitHub (May 6, 2025):

1.8.1 is live

@timothycarambat commented on GitHub (May 6, 2025): 1.8.1 is live
yindo changed title from [Question]: MCP `docker` commands on failing to start on Windows? to [GH-ISSUE #3719] [Question]: MCP `docker` commands on failing to start on Windows? 2026-06-05 14:46:19 -04:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
Desktop
Docker
Integration Request
Integration Request
OS: Linux
OS: Mobile
OS: Windows
UI/UX
blocked
bug
bug
core-team-only
documentation
duplicate
embed-widget
enhancement
feature request
github_actions
good first issue
investigating
needs info / can't replicate
possible bug
pull-request
Mirrored from GitHub Pull Request
 question
stage: specifications
wontfix
No Label Desktop bug investigating possible bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#2396
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?