github / github/copilot-cli

Stdio Docker MCP containers remain running after closing a session

Offen
#4,461 0 Kommentare 1 Reaktion 0 zugewiesene Personen Auf GitHub ansehen

Dieses Issue hat noch niemand übernommen.

area:mcp area:sessions
Vorherrschende Sprache
Shell
Sterne
11.2k
Forks
1.9k
Ø Merge
14 Std. 16 Min.
Gemergte PRs (30 T.)
6

Beschreibung

Describe the bug

When multiple Copilot CLI sessions use local stdio MCP servers running in Docker, closing one session does not terminate the Docker containers started for that session.

The behavior occurs with both:

  • A direct Docker MCP.
  • A Node wrapper that starts.

Because both MCP implementations reproduce the problem, this appears to be a Copilot CLI stdio MCP lifecycle issue rather than a problem in either Docker image.

After using multiple sessions, docker ps showed three active instances of each image:

matheusrhgestor/html2pdf-mcp
matheusrhgestor/html2pdf-mcp
matheusrhgestor/html2pdf-mcp
matheusrhgestor/kendo-ui-doc
matheusrhgestor/kendo-ui-doc
matheusrhgestor/kendo-ui-doc

In one reproduction:

  • A session was resumed around 11:45 local time.
  • Its session shutdown was recorded around 11:54 local time.
  • The kendo-ui-doc and html2pdf-mcp containers created around 11:44 remained in Up state afterward.
  • A new active session started another pair around 11:55.

Representative container output:

052529bdf996|matheusrhgestor/html2pdf-mcp|Up 7 minutes
1793cf8f1aad|matheusrhgestor/kendo-ui-doc|Up 7 minutes
9870e1fd5c7f|matheusrhgestor/html2pdf-mcp|Up 18 minutes
d1bc7dd02cd6|matheusrhgestor/kendo-ui-doc|Up 18 minutes
e161b8af2fb7|matheusrhgestor/html2pdf-mcp|Up 39 minutes
fc11f628ff81|matheusrhgestor/kendo-ui-doc|Up 39 minutes
Affected version

1.0.79

Steps to reproduce the behavior
  1. Configure the kendo-ui-doc MCP as a local stdio server using Docker:

    {
      "type": "local",
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "matheusrhgestor/kendo-ui-doc"
      ]
    }
    
  2. Configure the html2pdf MCP as a local stdio server using a Node wrapper that starts:

    docker run --rm -i --mount ... matheusrhgestor/html2pdf-mcp
    
  3. Start Copilot CLI session A and verify that one container per MCP is running.

  4. Start Copilot CLI session B and verify that a second pair of containers is running.

  5. Close session A using the normal session exit action, such as /exit or Ctrl+D.

  6. Run:

    docker ps --all --format "{{.ID}}|{{.Image}}|{{.Status}}|{{.Names}}|{{.CreatedAt}}"
    
  7. Observe that the containers created for session A are still running.

Image
Expected behavior

Closing a session should disconnect its stdio MCP processes and stop/remove only the containers created by that session.

The --rm option should remove each container after the corresponding docker run process is terminated.

Containers belonging to other active Copilot CLI sessions should remain running.

Additional context
  • Operating system: Windows 11 Pro
  • Architecture: x64
  • Shell: command prompt
  • MCP transport: local stdio
  • Docker images:
    • matheusrhgestor/kendo-ui-doc
    • matheusrhgestor/html2pdf-mcp

The direct kendo-ui-doc configuration uses docker run --rm -i.

The html2pdf wrapper uses Node child_process.spawn with stdio: "inherit" and also starts Docker with --rm -i. Therefore, --rm cannot clean up the container while the Docker process is still alive.

The issue is reproducible across more than one session and more than one MCP implementation. The likely failure is that closing a Copilot CLI session does not propagate termination to the local stdio MCP process and its Docker child process.

This appears related to:

https://github.com/github/copilot-cli/issues/4049
https://github.com/github/copilot-cli/issues/3440

This report concerns concurrent sessions and closing one session, rather than only /new or /resume.

Beitragsleitfaden

Beitragsleitfaden öffnen

Erste Schritte

  1. Lies das ganze Issue und danach den Beitragsleitfaden des Projekts.
  2. Schreib ins Issue, dass du es übernimmst — das erspart doppelte Arbeit.
  3. Forke das Repository und arbeite in einem Branch.
  4. Öffne einen Pull Request, der die Issue-Nummer nennt.

Rechercherichtung

Beginne damit, das Problem mit zwei Copilot CLI-Sitzungen zu reproduzieren, die lokale stdio MCP-Befehle verwenden, welche über Docker gestartet werden, und verfolge anschließend den Pfad zum Beenden der Sitzung für diese untergeordneten Prozesse. Als abgeschlossen gilt die Aufgabe, wenn das Schließen einer Sitzung nur deren Docker-Container beendet und entfernt, während die Container der anderen aktiven Sitzung weiterlaufen.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Bewertung

Tech-Stack
docker, node.js
Bereich
cli, devops
Issue-Typ
Bug
Schwierigkeit
4/5
Geschätzter Aufwand
3-5 Tage
Aktivitätsstatus
Ruhig
Klarheit
Größtenteils klar
Anfängerfreundlichkeit
52/100

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.