Most project tools added "AI" as a box that writes a summary of a task you already had. That is not how operators actually use models. They want an assistant that can create the task, put it on the right list, assign it, and log the hour — the same moves a person makes, without a person making them.
MCP is the product, not a slide
IterDone ships a built-in MCP server at /mcp. An API key with read and/or write scope authenticates it, the same key the REST API uses. Space-scoped keys stay space-scoped on both surfaces. The count is not a slogan: 89 tools covering spaces, folders, lists, tasks, comments, time, members, webhooks. If the UI can do it, the agent can do it.
Why 89 instead of a chatbot
A chatbot bolted onto a tracker still needs a human to click. Tools let the model take the action. "Create a list called Sprint 42, add these three tasks, assign them, set due dates" is one conversation against MCP, not a person reconstructing the board from a generated paragraph. The API keys and webhooks doc is the operator's version of this paragraph.
Headless is a first-class mode
You can run a workspace without anyone opening the app. Create the space, build the lists, close the tasks, export the JSON. Log in when you want to look. That is the point of shipping MCP and REST at the same depth as the UI, and it is why the Linear/ClickUp comparisons mention agents at all: IterDone vs Linear, IterDone vs ClickUp.
Try the surface
The live demo is the UI. The MCP endpoint is on a real workspace after signup — create an API key, point the client at /mcp, and list the tools. No per-member AI surcharge. The tools are the product.