Local (stdio) MCP processes leak — new processes spawned per session, never cleaned up

Description

Local MCP servers configured with "type": "local" (stdio transport) spawn a new process for every session opened, but the processes are never terminated when the session ends. Over time this leads to a large number of orphaned processes consuming significant memory.

Steps to reproduce

Configure a local MCP in opencode.json, e.g.:

"firefox-devtools": {
  "type": "local",
  "command": ["npx", "@padenot/firefox-devtools-mcp", "--firefox-path", "..."]
}

Open and close several sessions in OpenCode Desktop or the web UI
Check running processes: ps aux | grep firefox-devtools-mcp

Expected behaviour

When a session ends, the stdio MCP child process it spawned should be terminated.

Observed behaviour

Each session spawns a new npm exec + node process pair. After ~12 sessions, 24 orphaned processes were running consuming ~1.2GB RSS combined. They are never cleaned up until the host process is restarted or they are killed manually.

Plugins

opencode-pty, opencode-wakatime

OpenCode version

v1.3.0

Operating System

macOS (Apple Silicon)

Terminal

Ghostty

This issue might be a duplicate of existing issues. Please check:

v1.1.6 still has memory bloat - heap not released + MCP orphan processes #7261: MCP orphan processes accumulate across sessions after crash/exit (same symptom: orphaned MCP child processes not cleaned up, reproduces with ps aux | grep)
Per-session MCP server and LSP duplication multiplies memory usage across concurrent sessions #13041: Per-session MCP server duplication multiplies memory usage across concurrent sessions (same root cause: each session spawns its own MCP processes without sharing or cleanup)
[Bug] Web UI spawns duplicate MCP server waves during active session #9230: Web UI spawns duplicate MCP server waves during a single session (related: runaway MCP process spawning)

If your issue is distinct from these (e.g. specifically about the stdio/local transport teardown path rather than crash orphans or concurrent-session sharing), please clarify how it differs so it can be tracked separately.

Thanks for the references. Checking each:

v1.1.6 still has memory bloat - heap not released + MCP orphan processes #7261: Primarily about orphans after a crash and Go heap arena accumulation — this issue is about clean normal-exit sessions also failing to clean up stdio child processes.
Per-session MCP server and LSP duplication multiplies memory usage across concurrent sessions #13041: About duplication across concurrent sessions sharing the same project — this issue is about sequential sessions each accumulating orphans.
[Bug] Web UI spawns duplicate MCP server waves during active session #9230: Web UI spawning duplicate MCP waves mid-session — different trigger (something causes a respawn while the session is still active).

The specific case here is: after a clean session close (no crash, single session), the stdio/local-transport child processes are not terminated. The teardown path for type: local MCP connections doesn't appear to send SIGTERM to the child before the session is destroyed. Keeping this open as it's a distinct and more precisely scoped bug.

fix(mcp): close transport on failed/timed-out connections

Navigation Menu

Repository navigation

Local (stdio) MCP processes leak — new processes spawned per session, never cleaned up #19168

Description

Steps to reproduce

Expected behaviour

Observed behaviour

Plugins

OpenCode version

Operating System

Terminal

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Local (stdio) MCP processes leak — new processes spawned per session, never cleaned up #19168

Description

Description

Steps to reproduce

Expected behaviour

Observed behaviour

Plugins

OpenCode version

Operating System

Terminal

Activity

github-actions commented on Mar 25, 2026

athal7 commented on Mar 26, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions