The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Radtask MCP listing page.
Connect any AI to your task cockpit. RadTask is an AI task manager whose agent actually does the task — and asks first before anything irreversible. It runs a live MCP (Model Context Protocol) server, so Claude, ChatGPT, Cursor, or your own agent can create and track tasks that land in the user's RadTask cockpit, where a safe autonomous agent runs them.
Private beta. A tool with a safety design — not a guarantee.
Endpoint (Streamable HTTP, JSON-RPC 2.0):
Get your private token at radtask.ai/connect. The token is your cockpit's namespace — tasks an AI adds appear in your RadTask.
| Tool | What it does |
|---|---|
create_task | Capture a task in plain English (optional project) |
list_tasks | List tasks, optionally filtered by status |
complete_task | Mark a task done by id |
Most "task manager" MCP servers are queues — they track what an AI should do. RadTask does the task: a real agent researches, drafts, writes documents, and schedules in a sandbox — and a deterministic risk classifier holds only the irreversible steps (sending, spending, deleting, publishing) for the user's one-tap approval.
The safety posture is published and machine-verifiable:
Design goal: zero unsupervised destructive actions. Irreversible actions are always held for a human.
MIT © Sureel