RadTask MCP Server
RadTask is the AI task manager with a live MCP server. Connect Claude, ChatGPT, Cursor, or your own agent — it can create and track tasks in your cockpit, where a safe autonomous agent actually runs them and asks before anything irreversible.
Connect URL (Streamable HTTP, JSON-RPC 2.0)
https://radtask.ai/api/mcp
Auth header (preferred)
Authorization: Bearer YOUR_TOKEN
A token in a URL ends up in server logs, browser history and referrer headers. Sending it as a header keeps it out of all three.
Query form (still supported)
https://radtask.ai/api/mcp?token=YOUR_TOKEN
Not deprecated — every existing connection keeps working. Use it for clients that cannot set a header.
Get your private token at radtask.ai/connect.
Tools
create_task— capture a task in plain English (optionally under a project)list_tasks— list tasks, filter by statuscomplete_task— mark a task done
Claude Desktop / Cursor config
{
"mcpServers": {
"radtask": {
"type": "http",
"url": "https://radtask.ai/api/mcp",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}
If your client cannot set headers, use "url": "https://radtask.ai/api/mcp?token=YOUR_TOKEN" with no headers block — that form is still supported.
Why it's different
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 the safety valve holds anything irreversible for your one-tap approval. The published agent-safety.json is a machine-verifiable contract.
Machine-readable: /.well-known/mcp.json · /llms.txt · /for-agents · RadTask is in private beta — a tool with a safety design, not a guarantee.