Skip to main content
A prompt is a saved, step-by-step procedure that you start yourself. You fill in a few values, such as a state and a date, and your AI assistant follows the procedure instead of improvising one. Prompts are different from sample prompts. A sample prompt is a question you type in plain English. A prompt is a built-in procedure that calls several tools in a set order.

How to start a prompt

In Claude Code, prompts appear as slash commands. The format is:
Type /mcp__ste__ and let autocomplete list all seven. The ste part is the name you gave the STE MCP when you connected it, so use that name if it’s different, such as stesymmetry. In Claude Desktop, choose a prompt from the + menu.
  • Enter values in the order shown in the table below.
  • You can leave optional values off the end.
  • To skip an optional value in the middle, enter an empty "" in its place.
  • Prompts take identifiers and dates only. They never take wages or year-to-date amounts. If a procedure needs a payroll amount, your assistant asks you for it.
Starting a prompt is free, but the work it triggers isn’t. The calculations and lookups a prompt tells your assistant to run are billed like any other call. ste_rate_vs_paycheck runs two pay calculations.

The seven prompts

Enter dates as YYYY-MM-DD. A date in another format doesn’t cause an error. It matches nothing.

Examples

ste_release_impact

You get a change list that includes future-dated rows. To see what a change costs an employee, the prompt sends you to ste_pay_calc_diff.

ste_year_end_changes

You get the full total for the window, not the length of one page. It covers changes loaded as of today’s reference data, so a later release can add a change dated inside a window you already checked.

ste_rate_vs_paycheck

You get the difference in tax between two pay calculations that differ only by pay date. Your assistant asks you for any payroll amount it doesn’t have.

ste_engine_vs_reference

Both sides are pinned to the same release, so you see the tax and not a version mismatch. You get a comparison, or a statement that the two sides read different data.

ste_jurisdiction_dossier

Each tax is labeled levied, repealed, or computed inside the engine, with its effective date and the rate’s unit.

ste_reconcile_cents

The prompt checks the rounding rule, supplemental method, and bracket basis before it concludes the rate is wrong.

ste_benefit_treatment

Give either tax IDs, or a state and a tax type, but not both. If the reference tables have no explicit rule, the answer says so and asks the engine instead of guessing.

Which clients support prompts

Support for the tools other than Claude Desktop and Claude Code comes from each tool’s own documentation. Other tools may start prompts differently from the commands on this page, so check your tool’s MCP documentation. If your organization connects through a gateway or relay, the gateway must also pass prompts through. Some pass tools only, so you may see no prompts even in a supported client.

Next steps

Resources

Reference documents your assistant can look up.

Sample prompts

Plain-English questions by role, ready to copy.