Wingman Cloud is live: plans that sync across Claude, ChatGPT, and mobile

I shipped Wingman Cloud today and I need to write this down before the adrenaline fades.
Quick background: Wingman is an open-source MCP server that renders a persistent, interactive plan panel inline in your AI chat. You and the model both edit it. You click checkboxes from the panel. The model calls tick_task when it finishes work. State persists across conversations and restarts.
That was v0.1 and v0.2. Local, stdio, desktop only.
Today is different.
What shipped
HTTP/SSE transport for the cloud path. OAuth 2.1 with WorkOS AuthKit. Multi-tenant Postgres on Neon with user_id scoping on every query. Deployed on Render at https://wingman-mcp.onrender.com.
Same 12 model-visible tools. Same panel. Same prompt templates. Transport-selectable entry points in the same codebase: local users get stdio + SQLite, cloud users get HTTP/SSE + Postgres + auth.
The panel also got a full redesign: notched progress tape with a count-up percent tween, per-row status rails, a diff-based render reconciler keyed by task id so ticks and deletes animate without a full repaint, in-panel export sheet, and press-and-hold to reorder.
What this unlocks
Mobile Claude support was impossible over stdio. Claude mobile only reaches hosted servers over HTTP. Now it works. ChatGPT compatibility too. And plans sync across every device you have.
What stayed the same
The local version is not going anywhere. pip install wingman-mcp, configure stdio transport, zero telemetry, full anonymity. That commitment is in the README and I mean it.
One ask
I am on starter compute and storage. Connect at wingman-mcp.onrender.com in your LLM of choice and try saying "test case." Traction is what keeps the lights on and what justifies scaling this properly.
Open source. MIT. Built in public.