UGCS Farm, natively in Claude.
Turn any brand URL into evidence-backed UGC ads — without leaving Claude.
mcp.ugcs.farmClaude Web & Desktop MCP
UGCS Farm as a custom connector — one click and it's in your Claude account.
Six categories, 70+ tools
Every tool is granular — Claude never has to dump the whole analysis at once. It picks the exact slice for your question and cites its sources.
Scrape & discover
Pull live posts + ads from TikTok, Instagram, and Meta / TikTok / Google Ad Libraries.
- · scrape_facebook_ads_for_company
- · scrape_tiktok_videos
- · search_facebook_ads
- · discover_competitors_from_url
Decide & rank
Score patterns by commercial signal, filter sources by relevance, rank creatives by goal.
- · find_proof_grade_sources
- · rank_creative_patterns
- · explain_why_pattern_matters
- · get_hero
Read the audience
Classify comments by buying intent. Reach is a lie — signal is real.
- · analyze_engagement_quality
- · compare_creative_audiences
- · analyze_creator_audience_fit
- · compare_creators
Benchmark the category
Cross-corpus stats: which hooks / formats / CTAs actually survive in your category?
- · get_cta_benchmark
- · get_hook_benchmark
- · get_format_benchmark
- · get_pattern_signature_benchmark
Ship a production pack
Turn a winning pattern into a hook, script, storyboard, creator brief, and test plan.
- · get_ad_pack_summary
- · get_ad_pack_script
- · get_ad_pack_creator_brief
- · create_ab_test_plan
Cross-reference
Ad Libraries don't expose comments. We bridge paid → organic → real buyer questions.
- · find_organic_versions_of_ad
- · fetch_comments_for_paid_creative
- · preview_creative
- · record_ad_pack_feedback
What it looks like
One prompt (“what videos should I make for CalTrack”) — four tool-driven steps. Every step is a real MCP tool call; every claim traces back to scraped data.
Discover the niche
Claude calls scrape_tiktok_videos, search_instagram_hashtag, and returns a grid of high-performing creators in your category — handles, thumbnails, and view counts, straight from the platforms.
Diagnose weak audience
Not every viral video is worth copying. analyze_engagement_quality classifies every comment — when only 12% show buying intent, we flag it as a vanity-metrics creator. Reach without signal.
Confirm the winner
The one that matters: 46% of comments discuss the product, saves-to- likes ratio spikes, comments ask “what's the app name?”. This is the audience your brand actually converts.
Adapt the script
Claude turns the winning creative into a scene-by-scene remix — “The contradiction”, “The doubt”, first-frame direction, everything you need to hand a creator.
Example prompts
Once installed, drop any of these into Claude — the connector picks the right tools automatically.
- Chat
"Pull boAt's live Facebook ads, then for the top 3, cross-reference to their organic Instagram posts and pull the real buyer comments."
- Decision
"For a supplement brand, rank the top creative patterns for conversion. Explain the observed vs inferred evidence for #1."
- Audience
"Compare these three TikTok creators' audiences for a skincare brand. Whose followers actually ask buying questions?"
- Ship
"Build an ad pack from the winning pattern in this report, then generate an A/B test plan against our current baseline creative."
FAQ
What data does the connector access?
Only what you ask it to. All scrape data comes from public endpoints (TikTok, Instagram, Meta / TikTok / Google Ad Libraries). We don't touch your private accounts or credentials.
Is it free?
The MCP surface itself is free. Upstream API calls (ScrapeCreators for scraping, Gemini for LLM classification) are billed on our end at cost — see the pricing page for tier limits.
Does it work in other IDEs?
Yes — anywhere MCP works. Same config format for Cursor, Continue, Zed, and the CLI. The remote endpoint is https://mcp.ugcs.farm/mcp.