Skip to main content
Prompt Studio proposes prompts to monitor. This guide generates a pool of suggestions from an MCP assistant, reviews them with you, and accepts the selected ones as prompts. Generation uses no customer credits. Accepting monitored prompts adds recurring credit use, and can start one collection at the next scheduler pass.

Before you start

  • An active Business, Agency Business or Enterprise plan.
  • A connection with read:prompts and write:prompts, and a Manager or Owner role. See MCP write access.
  • For a topic generation, use the UUID of an existing topic. Looking it up with list_topics requires read:topics. A workspace-wide generation uses no topic ID.

Steps

Use your topic UUID and the suggestion IDs returned by get_prompt_suggestions in place of the synthetic values below.
1

Start a generation

The assistant calls generate_prompt_suggestions. Use source: manual for the whole workspace, or source: topic with a keywordId:
A workspace can start 3 workspace generations and 10 topic generations over a rolling 24 hours, with at most one workspace generation and one per topic running at a time. Reuse requestId for a lost response; change it for a new generation, including after FAILED.
2

Wait for it to finish

The assistant polls get_prompt_suggestion_run with the returned ID as suggestionRunId. On FAILED, stop and report the failure. Start another generation with a new requestId only if requested. On COMPLETED, continue. A client that also uses the REST API can subscribe to prompt_suggestions.ready and prompt_suggestions.failed instead of polling. The event carries the generation ID in data.runId.
3

Read the suggestions

get_prompt_suggestions returns the 20 recommendations by default, with a score, a reason and a tracked flag. Pass limit, up to 500, to read more of the pool; total gives its size. Use focus to reorder by funnel stage: DISCOVERY, CONSIDERATION, PERCEPTION or AUTHORITY. Skip suggestions that are already tracked, which means active and monitored. A suggestion equivalent to an archived or paused prompt reads false, and accepting it reactivates that prompt. Before selecting IDs, compare the returned run.id with the generation you followed. If a newer generation has replaced it, review the new pool separately.
4

Choose and preview

Pick up to 20 suggestions. The assistant calls accept_prompt_suggestions with dryRun set to true, or without a token:
Qwairy returns the outcome of each suggestion, the recurring credit use of the monitored prompts in credits.dailyChange, the cost of the collection the next scheduler pass would start in credits.oneOff, and a confirmationToken.
5

Confirm

If both amounts are acceptable, confirm. The assistant repeats the call with the same business arguments and the token, with dryRun omitted or false. The token is single use, valid for 10 minutes, and rejected if the arguments or the cost changed. Your MCP client may ask for its own approval too.
6

Check the prompts

Read the accepted prompts with get_prompts and check topic, monitoring and effective frequency. Acceptance has no frequency parameter. For a weekly cadence, use update_prompt with frequency: WEEKLY and the prompt’s current expectedVersion. Check any cost preview and read the result again.

Example conversation

Limits

  • Suggestions are ideas based on your workspace. They do not predict what an AI assistant will answer.
  • Accepting uses suggestions of the latest generation. IDs are bound to content and taxonomy: read them again after regeneration, and never construct them yourself.
  • A suggestion can be created, already tracked, not found or invalid: check every item.
  • Accepting starts no collection itself, but on a LIVE workspace it can make new or reactivated prompts due at the next scheduler pass. That collection uses credits once; the preview shows the amount in credits.oneOff.