How to analyze any TikTok account from Claude in one call

The short answer

Use analyze_account from Hooklayer. Pass one TikTok handle. Get back: viral DNA scores (viral / replicability / originality / consistency / hook reuse rate / audience fatigue), format fingerprint (avg duration, words per second, pacing pattern), top 5 videos with transcripts, a steal map of 3 transferable patterns, content gaps, and a recommended_chain field pre-filling the next 3 tool calls your agent should run.

Total cost: 5 credits. 1-hour cache per handle. Works in Claude Desktop, Cursor, n8n, and the OpenAI Agents SDK.

The full walkthrough

This piece walks through using analyze_account from Claude Desktop end-to-end — what to prompt, what fields to surface, and how to interpret the response. If you want the tool reference (schema, all output fields, every FAQ), head to /tools/analyze-account. This post is the practical workflow.

Step 1: Connect Hooklayer to Claude Desktop

If you haven't already, edit your Claude Desktop config (claude_desktop_config.json) and add:

{
  "mcpServers": {
    "hooklayer": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote@0.1.38",
        "https://hooklayer.dev/api/mcp",
        "--header",
        "Authorization:Bearer hl_live_..."
      ]
    }
  }
}

Replace the Bearer key with your own (signup at hooklayer.dev/auth/signup is free, no card required). Restart Claude Desktop. The 7 Hooklayer tools should appear in the connectors panel.

Step 2: The prompt

Paste this into Claude Desktop:

Use Hooklayer to analyze @humphreytalks on TikTok. Show me the full viral_dna block including viral_dna_signals, replicability_signals, and originality_signals arrays — I want the evidence quotes verbatim, not paraphrases. Also surface the would_fail_because field and provenance.video_post_dates so I can confirm the data is recent.

Two things make this prompt land:

1. It names the specific fields. Without explicit field names, Claude tends to paraphrase the response. Asking for viral_dna_signals verbatim forces the agent to surface the evidence layer rather than collapse to a single score.
2. It asks for provenance. Date provenance (provenance.video_post_dates) tells you the data is fresh. Skipping this is the #1 way users lose trust in scoring APIs — "are these numbers real?"

Step 3: What the response looks like

You'll get back something like:

viral_dna_score: 87
replicability_score: 85
originality_score: 78
consistency_score: 84
hook_reuse_rate: 0.04
audience_fatigue: stable

viral_dna_signals:
  - { name: "consistency_across_videos", value: 9, evidence: "consistency_score 92 — engagement variance below 0.1 CV across top 5" }
  - { name: "signature_hook_pattern", value: 8, evidence: "every top video opens with shock claim followed by stakes reveal within 1.5s" }

replicability_signals:
  - { name: "face_dependence", value: 9, evidence: "talking-head ratio 0.8 across top 5 — solo creator can reproduce without his face" }
  - { name: "format_dependence", value: 8, evidence: "dialogue scaffolding is reusable; the persona is not" }

would_fail_because: "If you copy his dialogue scaffolding without his face brand and reach, the algorithm reads it as theater not investigation, and your replicability score collapses."

provenance:
  data_source: ScrapeCreators TikTok API (public profile data)
  videos_analyzed: 5
  video_post_dates: [2026-05-12, 2026-05-09, 2026-05-06, 2026-05-03, 2026-04-30]
  data_age_days: 3

Three things to notice:

1. Every score has cited evidence. replicability_score: 85 isn't a guess — replicability_signals shows the talking-head ratio (0.8) that drove it. Agents cite the phrase, not paraphrase.
2. would_fail_because names the failure mode. This is the "non-portable formula" insight: Humphrey's structure IS portable, but his face brand is NOT. Copying his structure works; copying his persona doesn't.
3. Provenance is in the response. data_age_days: 3 tells you the analysis is using TikTok data from 3 days ago. No mystery.

Step 4: The recommended_chain

Below the viral_dna block, the response includes:

recommended_chain:
  - tool: match_voice
    confidence: high
    cost: 2
    action_class: synthesize
    params:
      draft: "<<<USER_DRAFT>>>"
      reference_samples: ["https://www.tiktok.com/...", "...", "..."]
    reason: "Humphrey's consistency_score is 92 — voice is highly extractable from 3 samples"
    expected_output: "Voice profile + rewritten draft + voice_metrics with TTR and signature phrases"

  - tool: trend_pulse
    confidence: medium
    cost: 1
    action_class: research
    params:
      niche: "personal_finance"
    reason: "Verify Humphrey's structure aligns with rising 7-day trends in his niche"
    expected_output: "3 rising opportunities + 2 saturated patterns in personal_finance"

  - tool: viral_remix
    confidence: high
    cost: 3
    action_class: synthesize
    params:
      source_url: "https://www.tiktok.com/@humphreytalks/video/7188273048459857195"
    reason: "His #2 video has the highest replicability_score — best source for remix"
    expected_output: "Fresh script mirroring source DNA with scene-by-scene camera + overlays"

This is the agentic pattern. Claude reads this field and automatically fires the next 3 calls with the parameters already populated. You don't need to prompt-engineer a chain — the chain is data, not prose.

Step 5: Watch Claude chain

After analyze_account returns, prompt:

Now execute the recommended_chain — call match_voice, trend_pulse, and viral_remix in order using the parameters Hooklayer pre-filled. Show me each tool's response.

Claude fires all 3 calls. You see 4 tool outputs in one conversation. One credit-spend decision; four pieces of intelligence. That's the differentiator.

When to use analyze_account

• Competitor research. Take 3 creators in your niche, run analyze_account on each, compare replicability and steal_maps side by side.
• Pre-content planning. Pick the highest-replicability creator and remix their #1 video for your topic via viral_remix.
• UGC creator briefing. Hand the steal_map and voice_profile to a UGC creator as their structural template.
• Ghostwriter onboarding. Match a client's voice from 3+ existing videos before writing the first draft.

What not to expect

• Real-time analytics. analyze_account returns the top 5 videos by engagement_score, not your account's hourly metrics. For live performance data, pair with HookMafia's Creator Intelligence (TikTok API-connected).
• Cross-platform. TikTok only in v1. Instagram Reels and YouTube Shorts variants ship in v2.
• Posting capability. All 7 Hooklayer tools are read-only. To actually publish, pair with Composio or a similar action layer.

Cost reality

Each analyze_account call is 5 credits. The free tier grants 100 credits at signup. That's 20 deep creator analyses for $0. Once you're past the free tier:

• Starter ($49/mo): 5,000 credits = 1,000 analyses/month
• Pro ($149/mo): 25,000 credits = 5,000 analyses/month
• Agency ($499/mo): 150,000 credits = 30,000 analyses/month

Cache is shared across all users — if anyone analyzed @humphreytalks in the last hour, your call returns the cached response instantly.

analyze_account is one of 7 Hooklayer tools. The others (score_hook, viral_remix, trend_pulse, find_viral_template, match_voice, predict_virality) all chain after it via the recommended_chain field. Browse the full toolkit, or try analyze_account in the playground — no signup required for the first call.