cli: mcp list reports "Request timed out" for healthy, connected servers #52563

Summary

opencode mcp list reports failed: Request timed out for 10/11 configured MCP servers, while the running server process has those same servers connected and healthy (verified via server log and live tool calls). The lister's standalone probing appears to time out on everything except instant local stdio servers.

Environment

• opencode version: 2.0.21
• OS: Darwin 25.2.0 (macOS, arm64)
• Terminal: Apple_Terminal, TERM=xterm-256color, COLORTERM=truecolor
• Shell: /bin/zsh
• Install/channel: latest (installed via https://opencode.ai/v2/install)
• Active plugins: one local file plugin auto-loaded from ~/.config/opencode/plugins/

Reproduction

1. Configure several MCP servers in ~/.config/opencode/opencode.jsonc (mix of remote HTTP and local uvx/binary stdio servers).
2. Start the TUI; observe the server log: mcp connected server=context7 tools=2, mcp connected server=gitmcp tools=5, plus local uvx servers connecting in 2–3s.
3. From another shell, run opencode mcp list.
4. Optionally confirm reachability at the same moment: curl https://mcp.context7.com/mcp → HTTP 405 (~0.6s), curl https://gitmcp.io/docs → 200.

Expected Behavior

mcp list reflects the live connection state (or at least completes an accurate probe): connected servers show connected, tool counts shown.

Actual Behavior

opencode mcp list output (11 configured servers; non-public server names redacted):

✗ context7             failed: Request timed out
✗ gitmcp               failed: Request timed out
✗ <remote-https-3>     failed: Request timed out
✗ <remote-https-4>     failed: Request timed out
✗ <remote-https-5>     failed: Request timed out
✗ <remote-https-6>     failed: Request timed out
✗ <remote-https-7>     failed: Request timed out
✗ <remote-sse-8>       failed: Request timed out
✗ <local-uvx-1>        failed: Request timed out
✗ <local-uvx-2>        failed: Request timed out
✓ <local-binary-1>     connected

Every remote server AND every uvx-based local server reports a timeout, while only an instant local stdio binary reports connected. At that same moment, the long-running server process (opencode serve --service) had 6 of these servers connected — public ones included (context7/gitmcp were reachable via curl from the same shell at the same moment).

--print-logs --log-level debug mcp list prints no log lines at all — only the status table — so the timeout cause is opaque. (--print-logs notes that server logs require --standalone.)

Additional Context

• The failures are uniform ("Request timed out") across HTTP-remotes and local uvx spawns, suggesting a systemic issue in the lister's standalone connection path rather than per-server config.
• Config was auto-migrated from v1 (mcp.servers, disabled: false normalized) and parses cleanly (opencode debug config exit 0).
• Reproducible: identical output across repeated runs minutes apart.