io.github.khan-ashifur/hooklayer

Analyze a TikTok, YouTube, or Instagram creator by handle or channel ID. Returns viral DNA scores (viral_dna_score, replicability_score, originality_score, consistency_score, audience_fatigue), content patterns, format fingerprint, top recent videos with transcripts, content gaps, a headline_insight object (the single largest quantified performance gap across length, hook, format, and cadence, plus its so_what), and suggested next research steps. Use when the user asks to analyze a creator, account, channel, or competitor. Supports TikTok (full transcript extraction), YouTube (Shorts and longform analysis with captions when available), and Instagram (Reels with best-effort transcripts).

Deeply validate ONE TikTok Shop product before committing money or content to it. Give it a product_id from product_scout (or any TikTok Shop product URL) and it reads the listing, samples reviews across the full depth rather than only the most recent page, pulls comparable competitors, and returns what buyers actually say. You get: provider-reported sold / review / rating / price evidence; buyer voice split into what buyers praise and what they complain about; unmet needs and purchase objections in buyers' own words; return and fulfilment signal; a Why Now read; competitor intelligence and where rivals are weak; risks separated into product risks, market risks and data-confidence risks; and a 'how to beat this' synthesis of concrete differentiation angles traced back to specific review evidence. Every section states whether it resolved - provider facts, HookLayer analysis and genuinely unavailable evidence are distinguishable, and thin evidence is reported as thin rather than filled in. There is no revenue or GMV output and sold counts are never multiplied by price. Use product_scout first to find candidates, then this to decide between them. Cost: 10 credits. Requires the shop_intelligence entitlement.

Turn a brand brief that just landed into a one-page creative blueprint the manager can forward to the creator and to the brand contact — hook + template + hashtag combo + trend-velocity check + shoot instructions, in one call. Use when a brand sends a product and needs content on a tight turnaround (e.g. "product arrived, need content in 48 hours") and the manager needs a defensible direction with no research time. Chains find_viral_template + trend_pulse + score_hook + predict_virality upstream but exposes them as one MCP tool so agents don't stitch them manually. Cost 7 credits (bundle discount vs firing the chain manually). Returns a blueprint object with verdict (GO | NEEDS_MORE_DATA | NO_GO) and verdict_reason, a hook (text + trend_still_alive: up-slope | plateau | fading | unknown), script, hashtags, shoot notes, risk flags, and manager talking points, plus a quality object (level + reason) describing completeness; when quality.level is not "full", or verdict is NEEDS_MORE_DATA or NO_GO, the blueprint is a starting point rather than a shippable direction.

Compare 2 to 4 saved analyze_product results side by side using only the stored evidence (identity, market proof, economics, competition, creator content, verdict). Read-only, 0 credits, no new analysis. Fields the source never reported are returned as not present, never estimated, and a freshness warning is included when the evidence ages differ by more than two days. Ids the user does not own come back in missing[]. Use after list_saved_research when the user asks which of their saved products to pursue.

Find proven viral templates in a niche with example videos. Returns templates ranked by performance, including hook patterns, format structures, average views, and example URLs. Use when the user asks what's working in a niche or wants concrete copyable structures. Supports the 18 canonical niches with optional angle narrowing via query parameter for more specific results (e.g., "postpartum strength" within Fitness).

Detect what changed since the last saved baseline for a tracked creator. Runs a fresh analyze_account under the hood, compares against the most recent CreatorSnapshot, persists the new snapshot, and returns ONLY the meaningful shifts (format, viral DNA, engagement, hook pattern, topic pillars, outlier videos). Use when the user asks 'what changed with @creator', 'anything new', 'check this competitor again', 'what's different since last time', 'check my tracked creator', or 'has anything changed'. Requires a previously tracked creator (call analyze_account + watch_account first if there is no baseline yet). Costs 5 credits per call because the fresh analysis is genuine (not cached). Returns a top_action pointing at the single next tool worth calling given the detected change. Returns status='no_meaningful_change' honestly when nothing shifted — no fabricated deltas.

Retrieve one saved result in full (the stored research payload, what was asked, timestamps) by its saved_id. Read-only, 0 credits. Only the authenticated user's own results resolve; any other id is not found. Use when the user wants to reuse, re-read or build on something they saved earlier instead of paying to run the tool again.

List the authenticated user's saved HookLayer research (newest first by default). Read-only, 0 credits. Filter by kind, search titles, choose sort newest|oldest|title, include archived, and page with the returned cursor. Returns summaries (id, kind, title, timestamps, dashboard link), not the full result; call get_saved_research for one. Use when the user asks 'what have I saved', 'show my saved products', 'find the analysis I saved for @handle'.

List every creator the authenticated user is tracking. Read-only, 0 credits. Returns each watch's id, platform, handle, when it was created + last updated, when the last snapshot was taken, and whether it's active. Does NOT return the snapshot payload itself — call watch_account (idempotent) to refresh a baseline. Use when the user asks 'what am I tracking', 'show my watches', 'list creators'.

Extract a creator's voice DNA from reference samples and rewrite a draft in their style. Requires at least 3 reference samples (video URLs or text). Returns voice profile (energy, humor, vocabulary, signature phrases), reusable prompt instructions, and the rewritten draft. Use when the user wants to write in another creator's style or match a specific voice.

Score a draft script for viral potential with adversarial verification. Returns a virality score, recommendation (ship/rework/no-go), viral DNA breakdown with evidence, attack vectors analysis, and calibration metrics. Use when the user has a finished draft and wants pre-publish verification. Pass either a script string or a video URL.

Find and rank current TikTok Shop / social-commerce product opportunities using evidence rather than opinion. Use it to answer questions like 'what skincare products are breaking out in the US under $40', 'show me products with strong reviews but low competition', or 'what is worth testing in this niche right now'. Returns products ranked by a deterministic 9-component engine (demand, momentum, creator_adoption, competition, saturation, review_strength, price_attractiveness, creator_concentration, freshness), each carrying an opportunity status (emerging | promising | crowded | mature | cooling | insufficient_data), a qualification record (what passed, failed and is unknown), a confidence band, evidence-cited why_now[], risks[], insufficient_signals[] and history_coverage. Status comes from a qualification gate, never from the score: emerging needs measured sold-count growth across 3+ dated observations spanning 7+ days plus an independent signal (review growth or creator adoption over time); mature means established demand without verified growth; insufficient_data means trajectory cannot be verified yet, however high the score. verification_summary.outcome is NO_VERIFIED_OPPORTUNITIES whenever no returned product passed the gate - say so plainly instead of presenting the list as opportunities, and use verification_summary.by_status to explain whether the category or the filters were the limit. Amazon competition is never measured (AMAZON_COMPETITION_UNVERIFIED); never describe a product as unsaturated on Amazon. The ranking is deterministic and contains no model output - read qualification, why_now, risks and components to explain to the user WHY something ranked where it did, and never present the score on its own. Provider-reported fields live under provider_reported.* and are null when TikTok did not expose them; null means unknown, never zero, and sold counts are never multiplied by price to imply revenue. There is no revenue or GMV field. To validate a singl…

Save a HookLayer result the user already received in this conversation so it appears under Saved Research in their HookMafia dashboard. 0 credits: this stores the result you already hold, it never re-runs the research tool. After a research result, offer once in one line to save it to HookMafia; call this only when the user says yes or asks to save, keep or bookmark a result. Pass the tool's structured result as `result` (billing and chain fields are dropped server-side), the kind that matches the tool, a short title, and a stable `request_id` so a retried save is deduplicated instead of duplicated. Returns the saved id and a dashboard link. Use list_saved_research / get_saved_research to read it back and compare_saved_products for 2-4 saved analyze_product results.

Score a TikTok, Reels, or Shorts hook against proven viral patterns. Returns a 0-100 score, percentile rank, matched pattern, strengths, weaknesses, and three improved hook variations. Use when the user has a draft hook to validate, wants to compare alternatives, or needs feedback before publishing.

Search for short-form videos by keyword across TikTok or Instagram. Returns up to 20 videos ranked by engagement, with view counts, likes, shares, comments, hashtags, author info, and URLs. Use when the user asks to find videos about a topic or keyword. Supports optional filters for niche, minimum views, recency window, and region. Pass product_id to check which videos link a specific TikTok Shop product: linkage is exact_product_link only when the video carries that product link; keyword_match videos are NOT verified for that product, and evidence_summary.product_linkage.status is VIDEO_LINKAGE_UNVERIFIED when none link it. Region counts as verified only from the provider-reported video region; otherwise it is unknown. View counts are cumulative and do not show current acceleration.

Research what is currently gaining traction in short-form content for a specific niche. Returns rising opportunities (formats, hooks, styles, topics) with growth signals, data sources, and saturated patterns to avoid. Use when the user asks what to post about, what's trending in a niche, or needs to validate a content idea against current trends. Supports the 18 canonical niches and optional region filtering.

Take a viral video and produce a fresh script that mirrors its structure and energy pattern for a new topic. Returns the extracted formula, scene-by-scene script with voiceover and visuals, camera directions, and text overlays. Use when the user finds a video they want to replicate the structure of. Pass either a video URL (TikTok, YouTube, or Instagram) or a transcript directly. When promoting a specific product, ALWAYS pass target_product + verified_product_facts so the generator does not fabricate product details.

Start tracking a creator so future calls can detect what changed. Creates a durable CreatorWatch + a baseline CreatorSnapshot capturing the creator's current viral DNA scores, format fingerprint, and top recent videos. Use when the user says 'track', 'watch', 'follow', 'keep an eye on' a creator/channel/handle. Idempotent — calling 'watch @same' twice with a fresh baseline in place returns the existing watch with credits_charged=0. Cost: 0 credits when a fresh existing watch snapshot or a compatible recent analyze_account cache can be reused. Otherwise a fresh baseline invokes analyze_account and costs 5 credits — this can happen even for an already-existing watch once freshness expires. Reuses the same 'analyze' permission as analyze_account — no new OAuth scope, no re-consent required.