26.6.2

MindMux v26.6.2 — Nightly

Highlights

• Run command tool & command allowlist — Agents can now execute shell commands directly from the chat interface. A configurable allowlist keeps you in control of which commands are permitted. Running tasks also show a read-only summary card so you can track what's happening at a glance.
• Model selector dropdown — A new dropdown in the sidebar lets you switch Claude models without leaving your session.
• Help menu with Send Feedback — A new Help menu in the menu bar gives you a direct path to send feedback to the team.

Fixes

• MCP tools no longer truncated — Lifted overly aggressive timeout caps on host-socket MCP tools, which were silently cutting off responses from longer-running tools.
• Task isolation choice now reflected correctly — The real isolation mode you selected is now threaded back through the create_task result, so the UI no longer misreports it.
• MCP import auth preserved — Fixed a regression where authentication settings were dropped when importing an MCP configuration. Auth is now correctly carried over.
• Expert output language guard hardened — Tightened the guard that enforces language constraints on expert-mode output.

Improvements

• Sidebar label casing normalized for consistency.
• Theme colors (secondary, muted, foreground) updated for better contrast across light and dark modes.
• MCP import validation messaging refined to be clearer when something goes wrong.