Skip to main content
These guidelines help AI assistants use rocketblue MCP safely and well. Agents can also call get_usage_guide to get this guidance as Markdown.
  1. get_organization_context: call once per session. It returns the plan, remaining credits, and every brand ID.
  2. get_brand_context: call before editing a brand. Use include to get topics, settings, and other sections in the same call.
  3. get_usage_guide: call when writing or restructuring prompts.

​
Use fewer calls

  • Tools that read a run accept brandId alone for the latest run. You do not need to look up the run ID first.
  • Batch every change set into one call: create_topics_and_prompts, update_prompts, update_outreach, and delete_items all take arrays.
  • update_brand_settings covers profile, competitors, and content settings in one call.

​
Regular analysis prompts

Regular prompts power your AI visibility reports. Write them like real search queries:
  • Neutral and unbranded. Do not put the client’s brand name in the prompt.
  • Consideration-stage when it fits. Comparison, evaluation, pricing, and “how to choose” intents reflect real discovery.
  • Organized by topic. Every prompt belongs to a topic.

​
Quota awareness

get_brand_context returns promptQuota.remainingSlots. Never create more prompts than the remaining slots allow. If the quota is full, delete unused prompts with delete_items first.

​
Generating prompts

  • generate_prompts with method: "ai" is free and saves about 8 prompts to the topic.
  • generate_prompts with method: "volume_research" costs 50 credits. It returns suggestions with real search volume but does not save them. Add the ones you want with create_topics_and_prompts and pass volume.

​
Perception prompts

Perception prompts are branded. They measure how models judge the brand on properties such as trust and clarity.
  • Use add_perception_prompts to create them and update_perception_prompts to change property or scoring logic.
  • Include goodScore and badScore so scores are easy to read.
  • Use correct_results to override a score that a model got wrong.
Custom perception prompts must be enabled on your plan.

​
Credits

Some tools spend organization credits. Each tool description states its cost.
  • Check creditsRemaining from get_organization_context first.
  • Tell the user the cost and get approval before running a credit tool.
  • Tools check the balance first and stop with a clear message if credits run out.

​
Destructive actions

delete_items deletes data for good. Deleting a topic also deletes its prompts and their results. Confirm with the user before calling it. Update tools are marked as destructive too, because they overwrite existing values.

​
Data collection

New prompts are queued on the brand’s latest run right away, so results start coming in shortly. Full reports still run on the brand’s schedule. nextEstimatedScheduledFetchAt from get_brand_context is an estimate based on schedule frequency and last activity.

​
Reading analysis data

​
Scope and permissions

  • Credentials are scoped to one organization.
  • Viewer and guest accounts cannot connect MCP.
  • With OAuth, actions are recorded under the user who connected. With an API key, they are recorded under the organization owner.
  • The rate limit defaults to 120 requests per minute per organization.

​
Example agent prompts

Connect an MCP client

Claude, Cursor, and generic HTTP setup.

Use cases

Digests, competitor watches, content pipelines.

Tools reference

Full tool list with costs.

Authentication

API keys and rate limits.