Taskriz is a developer platform for giving AI agents a way to ask people questions and receive structured responses. Developers can use plugins for Claude Code and ChatGPT, connect through Taskriz’s OAuth-protected MCP server, or call the HTTPS API from an agent, script, or workflow. The API lets developers register an agent with persistent polling or webhook delivery, exchange credentials for one-hour bearer tokens, and pair people through a 24-hour code, link, or QR code. Requests can contain interfaces written in a Smalltalk-based language with screens, lists, fields, switches, pickers, buttons, and file pickers; Taskriz runs those interfaces in a sandbox on the person’s device and returns button or field results as JSON. Agents receive replies and other events through polling, signed webhooks, or WebSockets, then acknowledge handled events. Delivery is at least once, so clients must de-duplicate by event ID; unacknowledged events are retained for 30 days. The documentation also covers catalogs of actions people can initiate, attachments up to 100 MB per file and 20 files per request, threaded conversations using correlation IDs, and recall of sent requests. MCP tools support agent creation, pairing, interface guidance and testing, sending requests, retrieving replies, attachments, catalogs, and recall. Requests count per recipient against plan limits, with the documented Free example allowing 500 monthly requests. Versioned `/v1` changes are additive, while breaking changes would use `/v2`.
