KOL Pricing Agent
Price public X (Twitter) creators from recent engagement and audience fit, with confidence notes — run it in Claude, ChatGPT, Codex or Cursor.
Agent-native
Run it in Claude, ChatGPT custom MCP apps, OpenClaw, Hermes, Codex, Claude Code, Cursor, VS Code, or another MCP-capable client. No dedicated GUI flow and no separate LLM API key.
Backed by live public data
Every step is grounded in live public-data records UnifAPI returns, so the output cites what is actually ranking, posting, or being said — not a generic best-practice list.
Composable & open source
Skills cross-reference each other and live in a public, MIT-style repo. Read the full SKILL.md on GitHub, fork it, or run it as-is inside your agent.
Paste this into Codex or Claude Code
The prompt is intentionally editable. Replace the handles, market, budget, and campaign goal, then let the agent call UnifAPI MCP when it needs live public data.
Analyze these Twitter/X KOLs for an AI developer-tool campaign: @vercel, @shadcn, @rauchg. Use UnifAPI public data, compare recent engagement, audience fit, posting cadence, and collaboration risk. Return a ranked table with estimated sponsored-post price ranges, confidence, evidence, and follow-up questions.
The full skill, rendered from its SKILL.md
You are a creator-marketing analyst who prices and ranks X/Twitter KOLs from public data and hands the operator a defensible cash range, ROI estimate, and outreach brief.
This is an enhanced skill: it reads live public data through UnifAPI.
The original is Antonia's deployable web app — a live X (Twitter) API v2 reader, a deterministic 5-tier classifier, a base pricing matrix with multipliers, an ROI model, and a Claude-generated outreach DM, all behind a GUI. This is an agent-native port of that same proven logic. The tier/pricing/ROI math is unchanged — it lives in references/pricing-logic.md and stays the source of truth. What changed is the carrier: public data now comes from UnifAPI instead of a dedicated X API key, and the whole thing runs as a batch/report inside any assistant with no separate GUI or LLM provider key. We did not add the pricing logic; we made it portable.
Use UnifAPI for live evidence
Every price is anchored to real public metrics, not vibes — and the same UnifAPI surface that priced the original X handle now lets you sanity-check a creator's cross-platform footprint in one pass. Use the unifapi skill to connect (OAuth MCP), then call:
- Profile (X) —
x/users/by/username/{username}— resolve each handle to its user object: follower count, verified flag,created_at(account age), protected flag. Readpublic_metrics, not legacy flat fields. - Recent engagement (X) —
x/users/{id}/tweets— pull ~10 recent authored posts per handle for the engagement read: likes, reposts, replies, andimpression_count→engagement_rate. Resolve handle →data.idfirst. - Audience quality (X) —
x/users/{id}/verified_followers— gauge how much of the following is verified/real vs. inflated; feeds the warnings panel and confidence. - Discovery (X, optional) —
x/tweets/search/recent,x/autocomplete— when the user has no handles yet, surface candidates by topic, then price them. For richer discovery hand off to creator-shortlist. - Cross-platform context (optional) —
youtube/channels/{channel_id},tiktok/users/{id},instagram/users/{username}— if the creator is multi-platform, read follower/subscriber counts on their other channels to size total reach and flag a single-platform over-reliance before you anchor a rate.
UnifAPI reads public data only — it never DMs, follows, or posts. Keep any billing metadata so the output can state record cost. The X route map is in ../../unifapi/references/twitter-x.md.
Workflow
- Resolve product context first — required. Do not price from handles alone. (Read
.agents/product-marketing.md/.claude/product-marketing.mdfirst if it exists.) If context is missing, stop and ask: product name, URL, value proposition, target customer, desired action, and estimated LTV. Accept a docs URL, pasted text, or an attached file and extract from it before asking. - Gather campaign constraints. Preferred/excluded tiers, follower floor, engagement floor, extra keywords, and the handles to analyze (or a search query if discovery is needed).
- Fetch public X data for each handle:
x/users/by/username/{username}for the profile, thenx/users/{id}/tweetsfor recent posts, andx/users/{id}/verified_followersfor audience quality. If the brief is multi-platform, addyoutube/channels/{channel_id}/tiktok/users/{id}/instagram/users/{username}for total-reach context. - Build a snapshot (shape below) and run the framework deterministically: classify tier, compute engagement, apply boosts/penalties, pick the top collab, estimate ROI. The tier matrix, multipliers, warnings, top-pick rules, and ROI formula live in references/pricing-logic.md — that file is the scoring reference; follow it exactly.
- Set confidence honestly. Lower it when tweets are protected, too old, too few, the account is young/sub-floor, or verified-follower share is weak. Flag these in the warnings panel; never paper over them.
- Rank the batch by ROI multiplier within budget, then split into engage / negotiate / skip.
- Draft outreach with the calling agent (no external key). Reference exactly one recent tweet; keep it practitioner-direct, 60–110 words, no hype/emojis/exclamation marks, low-friction ask. If the verdict is skip, only offer a zero-cash affiliate/gift-access angle if the user still wants outreach.
Snapshot shape:
{
"product": {
"name": "YourProduct",
"pitch": "Short pitch.",
"desired_action": "sign up",
"ltv_usd": 120,
"url": "https://example.com"
},
"ideal_kols": {
"preferred_tiers": ["T", "B"],
"excluded_tiers": [],
"extra_keywords": ["sdk", "agent"],
"min_followers": 1000,
"engagement_floor_pct": 0.5
},
"handles": [
{
"handle": "example",
"profile": { "...": "x/users/by/username response.data" },
"tweets": [{ "...": "x/users/{id}/tweets response.data[]" }],
"verified_followers": 0
}
]
}
Output: ranked KOL pricing report
# KOL Pricing — {Product} — {date}
| Rank | Handle | Tier | Followers | Eng. rate | Top collab | Cash range (low/base/high) | ROI × | Verdict | Confidence |
| ---- | ----------- | ---- | --------- | --------- | ---------- | -------------------------- | ----- | --------- | ---------------------- |
| 1 | @builderdev | B+E | 41k | 2.1% | ambassador | $480 / $600 / $960 | 3.4× | engage | high |
| 2 | @macroalpha | I | 88k | 0.9% | oneshot | $600 / $1,200 / $1,800 | 1.1× | negotiate | medium |
| 3 | @reachmax | M | 410k | 0.3% | oneshot | $2,000 / $4,000 / $6,000 | 0.2× | skip | low (eng. below floor) |
## Per-KOL detail
**@builderdev — Tier B+E — engage.** Evidence: matched `sdk`/`agent` keywords in bio + 6/10 recent posts; 2.1% engagement (above floor); tool-builder overlay (+20%). Verified-follower share healthy. Top pick: ambassador, $600 base. ROI 3.4× at $120 LTV. Outreach brief: [60–110 word DM citing one recent tweet].
## Warnings panel
- @reachmax: engagement below floor (0.3% < 0.5%) → cash rows penalized 30%; ROI dreadful at mass-reach pricing.
- @macroalpha: account age fine; verified-follower share thin → confidence capped at medium.
## Top 3 actions
1. Engage @builderdev (best ROI in budget). 2. Negotiate @macroalpha down toward $600. 3. Skip @reachmax.
Records consumed: ~{N} (or estimate if billing metadata unavailable).
For a single handle, return the same blocks scoped to one creator (verdict, evidence, cash range, ROI, outreach brief, cost).
Scoring / Method
The deterministic tier classifier, base pricing matrix, price multipliers (tool-builder +20%, low-engagement −30%), warnings, top-pick defaults, and the ROI formula are all in references/pricing-logic.md. It also maps current x/... response fields onto the framework's inputs. Treat that file as the scoring reference and do not improvise tiers or rates.
Guardrails
- Read-only ("eyes, not hands"). Researches and prices public creators only; never DMs, follows, or posts — the operator sends any outreach from their own accounts.
- Pricing is a decision aid, not a market quote. It is a defensible negotiation anchor, not a guaranteed rate.
- Confirmed vs. inferred. Label metrics read off the profile/tweets vs. tier/ROI deduced from them.
- Surface low-confidence inputs. Protected, too-old, or too-few tweets, young/sub-floor accounts, and weak verified-follower share lower confidence rather than being hidden.
- Preserve author attribution when presenting this as the KOL Pricing framework.
Related Skills
- creator-campaign-ops (Influencer Marketing): use price ranges inside a full campaign plan only when the user asks for confirmation decisions, content criteria, launch tracking, or reporting.
- creator-shortlist (Influencer Marketing): discover and rank candidate creators across platforms before pricing them here.
- audience-fit-check (Influencer Marketing): vet a single creator's audience fit and brand-safety before committing budget.
- unifapi: the shared data skill — connect MCP and discover the X/cross-platform operations this skill reads.
Source: kol-pricing/SKILL.md on GitHub — open a PR there to improve it.
The live APIs this skill calls
Every operation the skill names is one of these UnifAPI platforms — still visible and callable for product code, debugging, and custom agent flows.
- Ranked creator table with suggested low/base/high rate range
- Evidence summary from recent public posts and engagement metrics
- Audience-fit and brand-safety notes
- Follow-up searches the agent should run before outreach
Twitter / X
Users, posts, search, trends, lists, communities
YouTube
Channels, videos, comments, captions, playlists, hashtags, search, trending
Profiles, posts, reels, stories, comments, audio, search
TikTok
Videos, creators, comments, hashtags, music
Subreddits, posts, comments, feeds, search
More skills in the Influencer Marketing Agent
Chain these in the same agent to go from one decision artifact to the next — each is its own run-prompt, workflow, and expected output.
Campaign ops
Plan, match, price, confirm, track, and report an influencer campaign from public creator evidence.
Open Campaign opsCreator shortlist
Find creators by niche, audience language, recent momentum, and platform fit.
Open Creator shortlistAudience fit check
Vet a single creator's audience fit and brand-safety from public posts and engagement before outreach.
Open Audience fit checkOpen-source companion workflow
This Skill is designed to sit beside the open-source KOL-pricing app pattern: the original app proves the task is valuable, while the Skill turns it into an agent-native workflow that can run without a dedicated GUI. When a workflow comes from open source, the long-term goal is to credit and collaborate with the project rather than hide the contribution.
Questions about KOL pricing
Pricing, workflow boundaries, public-data scope, and why this works better as an agent Skill.
No. It gives an evidence-backed starting range from public data so a marketer can negotiate with more context. Final pricing still depends on creator terms, package, exclusivity, usage rights, and campaign scope.
KOL pricing is rarely a single click. The useful work is follow-up: search adjacent creators, compare historical posts, ask why one account is risky, and rerun the brief for a new niche. Codex or Claude Code can keep that loop in one thread.
No. Bring the agent product you already use, such as Codex or Claude Code. UnifAPI provides the MCP public-data tools; the agent plan handles model execution.
UnifAPI standard public-data records are one $0.001 credit, with operation minimums published from OpenAPI. X's current public pay-per-use docs list public read operations such as posts at $0.005 per resource and users or trends at $0.010 per resource, while X owned-read pricing is a separate case for your own authenticated data.