Skip to content

v0.9.0

Choose a tag to compare

View all tags
@samarkundal samarkundal released this 21 Sep 16:05
· 217 commits to main since this release
v0.9.0
2972385

Added

  • Connect triage from anywhere over HTTP: the daemon now serves its tools at POST /mcp, so Claude Code, Codex, or Cursor take a URL instead of spawning a process — http://localhost:5178/mcp, or :5188 in dev.
  • The workspace rides in the URL (?workspace=<id>). An id that matches nothing is refused at connect time with the list of ids that do, instead of quietly filing work into the default inbox.
  • An MCP tab in settings: the URL for the workspace you are looking at, and the one command or one file each client needs — Claude Code, Codex, Cursor, Codex Cloud (with what publishing an unauthenticated inbox actually means), and a stdio fallback.
  • npm run smoke:mcp covers both doors: the HTTP endpoint, and two concurrent chats.

Fixed

  • A chat no longer loses its triage tools when another session in the same workspace is live. One server instance was shared across the workspace, so the second session's connect was refused and its chat fell back to having no mcp__triage__* tools at all.
  • The init card stops reporting a shadowed server's status. A same-named entry in ~/.claude could show as failed next to tools that were working fine; a server now reads as working when its tools are in the session.
  • The stdio shim fails loudly and usefully: an unknown TRIAGE_WORKSPACE is refused rather than silently resolved to the default, and an unreachable daemon or wrong port says so instead of surfacing a raw fetch error.

Full changelog: v0.8.3...v0.9.0

Assets 2
Loading