Every few months OpenAI ships something that reads like a rebrand and turns out to be a real architecture change. The Apps SDK is one of those. "ChatGPT can build apps now" is easy to skim past — didn't ChatGPT already connect to tools via MCP? The honest answer: this is not a rename. It changes what an MCP server is allowed to send back, and if you run or evaluate a project management tool, the ChatGPT Apps SDK for project management is worth fifteen minutes of your attention.
Yes, it matters — but not the way the announcement headline made it sound. The Apps SDK doesn't replace MCP; it's built directly on top of it. It adds a way for an MCP server to return a rendered interface — a card list, a table, a small dashboard — instead of plain text, right inside the chat window. For project management specifically, that's the difference between ChatGPT describing your sprint board and ChatGPT showing it to you.
What is the ChatGPT Apps SDK, actually?
The Apps SDK is a developer toolkit, currently in preview, for building apps that render inside a ChatGPT conversation instead of opening a separate site or tab. It's built on the Model Context Protocol — the same open standard we've written about before — so if your tool already publishes an MCP server, the Apps SDK is an extension, not a replacement.
Here's the mechanism in plain terms. Your MCP server still exposes tools the way it always did: create a card, move a card, list what's in a column. What changes is the response. Instead of returning only text, a tool call can now return structured content with rendering metadata attached. ChatGPT reads that metadata and renders an actual interface component — built with standard web tech — inside the chat, synced to the conversation through what OpenAI calls the window.openai bridge. The official examples repo ships reference apps (a list view, a carousel, a map, a stateful shopping cart with OAuth) to show the pattern in practice.
Two things worth being precise about: it currently requires ChatGPT Developer Mode, which a workspace admin has to switch on, and it's still in preview — the API surface is expected to keep moving.
How is this different from the MCP connection you already have?
If you've connected Claude or ChatGPT to a project management tool before, you've already used plain MCP: the AI calls a tool, gets text or JSON back, and writes you a sentence describing the result. That still works, and it still ships. The Apps SDK sits one layer above it.
Plain MCP: you ask ChatGPT to "show me what's stuck in Review," and it replies with a written list of card titles and ages.
Apps SDK: the same tool call can return a small rendered table — sortable, maybe clickable — that appears inline, without you leaving the chat.
The protocol conversation underneath is identical. What's new is that a tool result can carry a UI, not just data. That's a genuinely useful capability for any tool where the shape of the information — a board, a calendar, a chart — matters as much as the words describing it.
What could the Apps SDK mean for project management tools?
This is where most explainers stop at "cool, widgets" and skip the actual implication for PM tools specifically, so here it is plainly.
Project management is one of the categories with the most to gain from this, because a board is inherently visual. "Three cards are overdue" is a sentence. A small rendered table with due dates and owners is the actual artifact a PM would want to see and act on — approve, reassign, comment — without a context switch to a separate app tab. If OpenAI's own examples (list, carousel, table-like views) are any indication, "render my sprint board inline in ChatGPT" is squarely the kind of use case the SDK was built for.
It also raises the bar for what "MCP support" means as a claim. A PM tool that exposes five read-only tools and calls that MCP support looks thinner next to one that can additionally hand back a real, interactive view of your board state. Buyers evaluating "AI + project management" tools in the next year should expect this distinction to start showing up in comparisons.
What this looks like for Comuna today
Here's the honest state of things, not the aspirational one. Comuna's ChatGPT integration today runs on plain MCP: connect via Settings → Apps → Developer Mode, and ChatGPT can create, move, and comment on real cards under its own identity, with the results described back to you in text. We have not shipped an Apps SDK widget for the ChatGPT side yet — the SDK is new and still in preview, and we'd rather say that plainly than imply otherwise.
What we do already have, inside Comuna itself rather than inside ChatGPT, is Canvas — AI-generated interactive HTML views that read live project data through a JS bridge and render as an actual dashboard or mini-app on the board, not a text summary. It's the same underlying idea the Apps SDK formalizes for ChatGPT: let the AI's output be a rendered interface tied to real data, not only a paragraph. Seeing OpenAI build a standard version of that pattern is, if anything, a sign the direction is right — we just built our version inside our own product first.
Where the Apps SDK falls short right now
Being clear about the limits is what makes the rest of this trustworthy.
It's still pull, not push. Nothing about the Apps SDK changes the fact that MCP execution happens when triggered — by you, in a chat, or by a scheduled prompt you set up. A widget rendering inline doesn't mean ChatGPT is watching your board and updating it live in the background.
Developer Mode and a paid plan are still required. Building or using Apps SDK experiences requires ChatGPT Developer Mode, enabled by a workspace admin, and the underlying Apps/MCP feature needs a ChatGPT Plus or Pro subscription — it isn't on the free tier.
OpenAI itself flags real security tradeoffs. Their own documentation warns developer-mode users to watch for prompt injection, model mistakes on write actions, and malicious MCP servers requesting more access than they should. A rendered widget doesn't remove that risk surface — if anything, richer interactivity is one more thing to audit before you grant write access to a tool.
It's preview software. APIs in preview move. If you're a PM tool builder deciding whether to invest engineering time now, that's a real cost to weigh against being early.
None of that makes the SDK unimportant — it's the clearest signal yet that "AI in your PM tool" is moving from a chat sidebar toward the AI rendering the actual tool inline. It just isn't finished, and anyone telling you otherwise is skipping the fine print.
FAQ
What does the ChatGPT Apps SDK mean for project management specifically?
It means a PM tool's MCP server can return a rendered view of your board — a table, a card list — instead of only a text description, right inside a ChatGPT conversation. The protocol connection is the same MCP you already use; what's new is that tool results can carry a UI.
Is the ChatGPT Apps SDK the same as MCP?
No. MCP is the underlying open protocol for connecting an AI to tools with read/write access. The Apps SDK is built on top of MCP and adds a way to render an interface from a tool's response. You can use MCP without the Apps SDK; the Apps SDK requires MCP underneath it.
Do I need a ChatGPT Plus or Pro subscription to use this?
Yes. Both the underlying Apps/MCP feature and Developer Mode require a ChatGPT Plus or Pro subscription — it isn't available on the free ChatGPT tier.
Does Comuna use the ChatGPT Apps SDK today?
Not yet. Comuna's ChatGPT integration runs on plain MCP today — real card creation, moves, and comments with full attribution, described back to you in text. We haven't shipped an Apps SDK widget, and the SDK itself is still in preview.
Comuna is free forever — no credit card, bring your own AI. Spin up a workspace and try it.