UGC Agent

Showing 25 of 32 from the live probe.

• Get Agent Context

  Return auth mode, plan, usage, preferences, service config, and operating instructions.

• Run UGC Agent

  Run a full UGC Agent sourcing/refine/export turn on the web app's search engine. Returns an app_url for table review and CSV export. Small targets finish inside the call and return the creators. A longer search returns status 'running' with list_id, found_so_far and an instruction: wait about 30 seconds and call get_run_status with that list_id instead of calling this tool again. Negations in the goal are honoured; plan limits, stored preferences and cross-run dedupe are applied by the web app f

• Research Company

  Research a brand/product/company so the LLM can derive niche and creator-search angles. Set for_collab_search=true when the user wants creators who ALREADY worked with this brand — also researches its official Instagram/TikTok handles and hashtags for find_brand_creators.

• Find Brand Creators

  Content-first search for creators who ALREADY made content for specific brands ('worked with Cal AI'): searches Instagram posts tagging the brand, brand hashtags, and TikTok keyword videos, and queues the post authors as candidates with proof. Brands' own accounts are excluded; stale posts (>18 months) don't count. ALWAYS follow with enrich_and_score passing the same brand names in `brands` — only verified collaborators are kept.

• Analyze Seed Creators

  Analyze example Instagram creators to derive criteria before sourcing similar creators.

• Discover Creators

  Search Instagram + TikTok for creator candidates and queue them for enrichment. Pass many varied queries per call; each query is searched on its own (they are never concatenated or replaced with a fixed niche taxonomy). Use similar_to to snowball off already-qualified Instagram handles.

• Enrich And Score

  Enrich queued candidates or specific Instagram handles, then keep only qualified on-niche creators with emails (unless the user's accept-no-email preference or an explicit 'emails optional' request relaxed the email requirement). Unless min_followers/max_followers are explicitly set, every mode uses the default micro-UGC band of 100–9999 followers with no automatic widening. The response includes applied_follower_bounds and follower_policy. Set exclude_agency_emails when the user asks for direct

• List Leads

  Return full kept lead rows for this MCP conversation.

• Remove Leads

  Remove kept leads by exact handle.

• Export CSV

  Return a CSV of kept leads plus an app_url for table review/export. Omit lead_ids to export all kept leads in the MCP conversation.

• Draft Creator Outreach

  Draft short, direct outbound copy (DM and/or email) for contacting a creator. Keeps it concise and never invents pay, proof, clients, or results. Use it after keeping leads to write personalized outreach.

• Analyze Creator Content

  Fetch recent posts for kept creators, compute per-creator virality (max views relative to followers), rank them, and return each creator's top videos with captions and hooks — transcribed from the actual video openings when possible, else caption openings. Use for questions like 'which of my creators have the most viral videos and what hooks do they use'. Results persist on the leads for 24h.

• Find Viral Trends

  Collect evidence-backed current TikTok/Instagram posts and grounded web sources, then return 1-5 validated trends with proof links, observations, and safe ways to use them. Use before ideate_ugc_content for trend-aware ideas. Evidence remains available when generation is partial or fails.

• Ideate UGC Content

  Generate structured UGC video ideas by crossing a library of proven short-form formats (problem→solution, before/after, unboxing, myth-busting, trend remix, etc.) with the brand/product/audience/pain points. Each idea has a hook, a 3-beat outline, why it works, and a CTA. The output is SCAFFOLDING: rewrite every hook and beat so it's specific to the product before presenting. Pass trend_context from find_viral_trends for trend-aware ideas.

• Draft UGC Script

  Draft a timed short-form UGC video script: hook (0-3s), body scenes with time ranges + shot directions + draft voiceover + on-screen text, CTA, caption, and hashtags. The voiceover lines are drafts — polish them into natural spoken language while keeping the timing structure. Never add claims or results the user didn't provide.

• List Automations

  List the authenticated user's lead-sourcing and trend-research automations with lifecycle state and recent status. Unsupported legacy records are excluded. Read-only; use this before editing or deleting an automation.

• Create Lead Automation

  Collect explicit values one missing field at a time in the returned order: criteria or Agent Preferences, lead_count, frequency, conditional weekday/day_of_month, exact daily_time, timezone, and an explicit recipient/account-email choice. Resolve 'local' or a city to a canonical IANA timezone before supplying timezone; never pass the literal 'local'. Once complete, creation writes immediately; the first run is scheduled for the returned next_run_at.

• Create Trend Research Automation

  Collect explicit values one missing field at a time in the returned order: topic, platform, frequency, conditional weekday/day_of_month, exact daily_time, timezone, and an explicit recipient/account-email choice. Resolve 'local' or a city to a canonical IANA timezone before supplying timezone; never pass the literal 'local'. Once complete, creation writes immediately; the first run is scheduled for the returned next_run_at. Region Global is fixed product behavior.

• Update Lead Automation

  Explicit two-phase write; never infer or default configurable values. Timezone is not an update input; the stored timezone is shown in the preview and preserved. Supply automation_id and at least one explicit changed value; untouched stored values are preserved and are not re-asked. A complete call without confirmation_token only previews the full resulting configuration and changed fields; repeat the identical call with the token after later user confirmation to write.

• Update Trend Research Automation

  Explicit two-phase write; never infer or default configurable values. Timezone is not an update input; the stored timezone is shown in the preview and preserved. Supply automation_id and at least one explicit changed value; untouched stored values are preserved and are not re-asked. A complete call without confirmation_token only previews the full resulting configuration and changed fields; repeat the identical call with the token after later user confirmation to write.

• Pause Automation

  Pause an owned automation without canceling active runs or pending deliveries. Future scheduling stops, nextRunAt becomes null, and the operation is idempotent.

• Resume Automation

  Resume an owned paused automation and schedule its next normal frequency-aware occurrence. It never triggers an immediate batch and the operation is idempotent.

• Run Automation Now

  Make an owned active automation due immediately for the Railway scheduler. It does not change the recurrence, create a run ID, or run synchronously; the next cron tick claims it.

• List Automation Runs

  List normalized run history for an owned lead-sourcing or trend-research automation. Defaults to 20 runs and accepts a maximum of 100.

• Get Automation Run

  Poll one owned automation run for status, progress, errors, delivery state, and trend results.