Prompts and resources
Besides tools, the server offers two other kinds of thing an MCP client can use: prompts, which are ready-made multi-step workflows a user can pick, and resources, which are reference material an assistant reads rather than calls.
Prompts
A prompt appears in the client as something the user chooses — in most clients, a named command — rather than something the assistant decides to call. Each one sequences several tools so the answer comes out complete instead of one query at a time.
| Prompt | What it does |
|---|---|
| analyze_work_order | Walks one work order end to end: header, positions, routing operations and material lines, flagging delays and component shortages. |
| production_status_overview | Builds a shop-floor status overview from the work-order KPI and status-breakdown tools. |
production_status_overview is built on the analytics tools, so it answers with counts rather than
by pulling every order. It stays fast on a database with a long production history.
Resources
A resource is readable reference material with a fixed address. An assistant can read one without spending a tool call, which makes them the cheapest way for it to orient itself.
| Resource | Address | What it holds |
|---|---|---|
| beas_usage_guide | beas://guide | How to use the server: the tool families, the two authentication schemes, and query tips. |
| beas_entity_catalog | beas://schema/entities | Every API entity set name reachable through beas_odata_query, as a JSON list. |
The catalog is generated from the API's own model, so it is never out of step with what the server
will actually accept. Pair it with beas_entity_schema, which describes the fields of one of those
entity sets — see Tools.
Examples
The assistant writes queries with field names that do not exist. It is guessing the schema
instead of reading it. Ask it to read beas://schema/entities for the entity set name, then call
beas_entity_schema for the fields. Both are cheaper than a failed query, and the field names come
back exactly as the API spells them.
Searches by name return nothing on a HANA database although the record exists. Text search is case-sensitive there. Match the stored capitalisation, or have the assistant search on a code instead of a description.