Qingflow MCP Assistant

Problem

When an agent has to work with Qingflow, the failure points are often operational rather than conceptual: whether qingflow-cli is present and current, whether the account is authenticated, whether MCP servers are connected, and whether record field IDs are resolved before CRUD calls. Qingflow also spans two domains: the user domain for records and tasks, and the builder domain for app schema, layout, flow, views, and charts. Calling tools without this setup can lead to wrong parameters, blocked terminal logins, or garbled output.

How It Works

The skill turns Qingflow requests into a constrained workflow:
- State checks: verify CLI update state, installation state, and login status.
- Authentication: if not logged in, ask for email and password, then pass --email and --password directly in agent contexts to avoid interactive blocking.
- MCP setup: when MCP tools are needed, configure qingflow-user for records, tasks, and import/export, and qingflow-builder for schema, layout, flow, views, and charts.
- Execution: prefer qingflow --json for machine-readable results; use MCP tools such as record_*, task_*, and app_* when connected. It emphasizes calling record_browse_schema_get_public before record_list, and record_insert_schema_get_public before record_insert_public, so field IDs and types are resolved first.

Boundaries and Notes

This skill fits WorkBuddy or MCP-capable agent environments, especially structured requests involving record lookup, task execution, and app flow maintenance. It does not create Qingflow accounts for users, nor does it cover unrelated office automation. If MCP transport errors, Python path issues, version mismatch, or Chinese terminal garbling occur, fall back to the documented troubleshooting path, such as writing --json output to a temporary file before reading it.