Workflow prompts
Use Creator Eagle's prompt templates for reliable brand audits, creator sourcing, and long-running job recovery.
Creator Eagle exposes three MCP prompts in addition to its tools. Prompts are user-selected workflow templates: retrieving one returns structured messages for the agent, but does not execute a tool, authorize paid work, or consume credits.
Prompt support depends on the client
Clients discover prompts through prompts/list and retrieve one through prompts/get. Some MCP clients and aggregators expose tools without exposing prompts. In that case, use the workflows on this page; the server instructions and individual tool descriptions still carry the critical call-specific rules.
brand-sponsorship-audit
Use this before creator lookalike sourcing when you need to learn which influencers already work with a target brand and its direct competitors.
Arguments:
| Argument | Required | Purpose |
|---|---|---|
brand | Yes | Target brand name or public website domain. |
competitors | No | Direct competitor names/domains, separated by commas or newlines. |
researchContext | No | CRM, email, brief, preference, and prior approval/rejection evidence already gathered. |
The returned workflow tells the agent to:
- Validate direct competitors and resolve every company with
find-brands. - Use
onboard-missing-brandonly when a legitimate brand has no exact match and a verified official domain, stop on collisions or blocked brands, then re-runfind-brands. - Record canonical IDs, fingerprints, readiness, asynchronous IDs, progress, and failures in a compaction-safe Markdown ledger.
- Start each explicitly authorized paid sponsorship search once, then reuse its IDs while polling.
- Wait for the target and every runnable competitor to become terminal.
- Retrieve every page of every brand roster only after those runs finish.
- Report all supporting evidence and confirm that it did not submit lookalike searches.
Selecting this prompt is not paid-processing approval. A request that explicitly asks the agent to run the complete/full sponsorship-detection flow supplies that approval; otherwise the agent must pause before start-brand-sponsorship-search.
creator-sourcing-workflow
Use this after any prerequisite research or approval phase is complete.
Arguments:
| Argument | Required | Purpose |
|---|---|---|
campaign | Yes | Brand/campaign name and desired sourcing outcome. |
brief | Yes | Audience, creator preferences, content, geography, performance, exclusions, and deliverables. |
seedCreators | No | Preferred YouTube seeds, separated by commas or newlines. |
antiExamples | No | Known poor-fit creators, separated by commas or newlines. |
The workflow initializes one influencer list, reuses its listId, profiles seeds once, submits intentional searches once, routes provider and Trigger identifiers correctly, and applies inexpensive filters before multimodal enrichment. It recommends screening promising channels with 3–5 recent videos first and spending deeper analysis or contact-discovery credits only on finalists.
resume-long-running-work
Use this after a callback wakes an agent, after a context compaction, or when work must continue in a later session.
Arguments:
| Argument | Required | Purpose |
|---|---|---|
references | Yes | Saved IDs together with the tool that returned each ID. |
objective | No | The terminal result the original run was meant to produce. |
The prompt routes each identifier through exactly one continuation path:
| Reference | Continue with | Terminal condition |
|---|---|---|
Trigger runId | poll-task | Completed output or failure |
Durable operationId | get-operation-status, then get-operation-result | succeeded, failed, or canceled |
Brand searchRunId | get-brand-sponsorship-search | terminal: true |
Similar-search provider jobId | poll-similar-creator-search | Enriched results, failure, or handoff to a Trigger runId |
Do not identify a reference by its format alone; retain the originating tool beside every ID. A callback is a wake-up signal, not a substitute for verifying terminal state and retrieving the result.
Why guidance appears in three places
- Server instructions carry the shortest always-on routing and authorization rules.
- Prompts carry complete multi-tool workflows that users can intentionally select and customize.
- Tool descriptions keep limits, prerequisites, and the next call beside the exact operation they govern.
This layering keeps capable clients efficient without making correct operation depend on prompt support.