Skip to content

Commit aef90cf

Browse files
omesserclaude
andcommitted
docs(settings): Say why OpenCode registration needs a restart
The steps were corrected on main while this was in flight, so the user-facing text is already right. What is missing is the reason, and without it the next reader has a bare assertion that OpenCode "does not re-read MCP mid-session". Record what was measured and where the wrong instruction came from. `anomalyco/opencode#6719` is titled "[FEATURE]: slash command for reload" and is open, so `/reload` was never a shipped command. Against OpenCode 1.18.30, a server written into the config of a running `opencode serve` never appeared in `GET /mcp` or `GET /config`. Cheap insurance against someone restoring `/reload` from the same source that put it there the first time. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
1 parent fb8bc78 commit aef90cf

1 file changed

Lines changed: 8 additions & 0 deletions

File tree

‎src-tauri/src/settings.rs‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -353,6 +353,14 @@ fn byo_registration(harness: &str, url: &str, token: &str) -> (String, String, S
353353
),
354354
token.to_string(), // Show token separately for interactive paste
355355
),
356+
// Restart, not `/reload`: OpenCode reads its configuration once at
357+
// start and caches it. A server written into the config of a running
358+
// session never appears — `GET /mcp` and `GET /config` both still
359+
// report the old set. `/reload` was in these steps because
360+
// `sst/opencode#6719` was read as a shipped command; it is an open
361+
// feature request. Telling the user to run it leaves them talking to a
362+
// session that still holds the previous run's port and token, with
363+
// nothing to show that anything failed.
356364
"opencode" => (
357365
format!(
358366
"opencode mcp add ai-buddy --url '{url}' --header \"Authorization=Bearer {token}\""

0 commit comments

Comments
 (0)