fix(settings-sync-daemon): force-overwrite Helix-owned context_servers

[lukemarsden]

Summary

Tail-end fix for the MCP-cache-contention investigation that PR #2418 addressed. After PR #2418 merged, the user reported chrome-devtools / drone-ci / github still showing "Context server request timeout" in Zed on a long-running spec-task container — even though the container was running the new helix-ubuntu:6de75e image (with all the global MCP binaries) and helix-api was running the new zed_config.go (with chrome-devtools pointing at /usr/bin/chrome-devtools-mcp instead of npx).

Root cause: ~/.config/zed/settings.json is a symlink to /home/retro/work/.zed-state/config/settings.json, which lives on the persistent ZFS-backed /home/retro/work volume. The settings-sync-daemon's mergeSettings deep-merge of context_servers lets USER entries win for every name, so the OLD command: "npx", args: ["chrome-devtools-mcp@latest"] written by the pre-PR-#2418 API last week was treated as a user customization and pinned forever, no matter how many times the new API generated the correct config.

extractUserOverrides had the symmetric bug: it captured the stale on-disk entry as a user override and round-tripped it back to the API.

Fix

New constant HELIX_OWNED_CONTEXT_SERVERS = {chrome-devtools, helix-session, helix-desktop} — the names hardcoded in api/pkg/external-agent/zed_config.go. Two corresponding behavior changes in api/cmd/settings-sync-daemon/main.go:

1. mergeSettings — skip user-side context_servers entries whose name is in HELIX_OWNED_CONTEXT_SERVERS so Helix's hardcoded definition unconditionally wins. Also strip helix-owned names from the "user-only" branch so a stale on-disk entry can't survive even when the API temporarily emits no context_servers.

2. extractUserOverrides — never capture helix-owned names as user overrides (otherwise the stale entry would round-trip back to the API and force the next sync to re-write the OLD value to disk, permanently nullifying the force-overwrite from Website #1).

User-configured MCPs (e.g. drone-ci, github, custom servers from project skills or app config) are NOT in the helix-owned set — those legitimately can be edited by the user in their on-disk settings.json and must round-trip.

Tests

• TestMergeSettings_HelixOwnedContextServersWin (4 sub-tests):
  • force-overwrite chrome-devtools when user has stale npx version
  • force-overwrite helix-session when user has stale session_id URL/token
  • user-configured drone-ci still wins (positive control — non-helix-owned MCP)
  • strips helix-owned names even when helix temporarily has no context_servers
• TestExtractUserOverrides_SkipsHelixOwnedContextServers (2 sub-tests):
  • stale on-disk helix-owned entries are not captured as user overrides
  • non-helix user overrides still round-trip (positive control)

Verified regression-test power: with both guards commented out (if false && HELIX_OWNED_CONTEXT_SERVERS[name] {), all 5 of the 6 force-overwrite/skip sub-tests fail with the diagnostic messages in the assertions — exactly the regression we're guarding against.

Test plan

• CGO_ENABLED=0 go test -v -run "TestMergeSettings_HelixOwned|TestExtractUserOverrides_SkipsHelix" ./api/cmd/settings-sync-daemon/ -count=1 — all 6 sub-tests pass.
• Disabled both guards, re-ran — 5 of 6 fail with the expected diagnostic. Restored.
• go build ./api/cmd/settings-sync-daemon/ clean.
• CI (Drone)

Follow-up after merge

The user's affected long-running container (spt_01kqc4ev5rt9rknk6g8dbkzj9a) needs a ./stack build-ubuntu + new session for the daemon binary to be replaced (per CLAUDE.md: settings-sync-daemon does not hot-reload). Once the new daemon runs, it'll force-overwrite the stale chrome-devtools/helix-session/helix-desktop entries on the next sync, and Zed will pick up the new /usr/bin/chrome-devtools-mcp config.

Related

• Builds on PR fix(mcp): pre-install MCP servers globally to skip npx cache contention #2418: fix(mcp): pre-install MCP servers globally to skip npx cache contention #2418 (npx-cache-contention + Fix 1a/1b/2)
• Builds on Zed PR reverse the color of the active session #56: fix(external_websocket_sync): defer UserCreatedThread until first user message zed#56
• Design doc: design/2026-05-13-mcp-cache-contention-and-duplicate-claude-spawn.md

🤖 Generated with Claude Code