Persuasion Taxonomy

com.coppicav1.0.0更新於 Oct 4, 2026

Plan and check marketing copy against the nine questions every reader asks. Free, no API key.

已驗證Streamable HTTP可網頁執行Data & Analytics Business & Commerce

概覽

AI 產生的概覽

讓助理依照讀者會問的九個問題來規劃、診斷與改進行銷文案，檢索 766 種說服技巧，並規劃或解讀 A/B 測試。

功能: 這個伺服器為助理提供讀者模型：每位讀者都會默默提出的九個問題。相關工具可依這些問題規劃行銷文案，透過指出文案回答了或漏掉哪些問題來診斷草稿，解釋文案為何無法轉換，並檢查標題與主張。它也能檢索從真實廣告整理出的 766 種說服技巧目錄，回傳附範例的完整條目，並規劃或解讀 A/B 測試結果。它本身不呼叫任何 AI，思考由你的模型完成。
適用情境: 當助理需要撰寫或審閱廣告、到達頁、電子郵件或銷售信，而你希望在發布前先做檢查時，值得加入。它也適合診斷有點擊卻沒有銷售的文案、尋找不那麼制式的建立信任方式，以及決定或解讀 A/B 測試。
執行需求: 透過 Streamable HTTP 連線至託管網址的遠端 MCP 端點；不需要登入、API 金鑰或環境變數。也可以用 npx 在本機執行。使用遠端方式時需要能連線到該託管端點。

安裝前請注意

伺服器聲明不保留你送出的內容：文案、簡報與測試數據只用於該次請求，之後即丟棄，託管方會像一般網站一樣短期保留請求記錄。它不寫入資料，也不涉及付款。使用託管端點時，你提交的文案與測試數據仍會離開本機。

安裝

在 SourceWeft 中

開啟儀表板中的 Persuasion Taxonomy，將其新增到工作區。
為需要使用其工具的對話啟用該服務。

Web executable，透過 Streamable HTTP。遠端服務在工作區中設定後即可從網頁執行環境執行。

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "persuasion-taxonomy": {
      "type": "http",
      "url": "https://taxonomy.coppica.com/mcp"
    }
  }
}

README

Persuasion Taxonomy MCP

Most AI-written marketing copy fails for the same reason. The model answers every question the reader has with the most expected move, and the most expected move is exactly what readers have learned to skim past.

This server gives your AI a better model of the reader. Everyone who reads an ad, a page or an email is silently asking nine questions:

Why am I even reading this?
Is this written for me?
How bad is my problem, really?
Am I thinking about this correctly?
Can I trust the claim?
What does my life look like after?
What's stopping me from saying yes?
Why should I act right now?
Do I like and trust who's speaking?

Copy persuades when it answers the questions its goal needs, answers them truthfully, and the answers fit together. The server helps your AI plan copy around those questions, check a draft against them before anyone sees it, and reach past the obvious move for one of 766 persuasion techniques documented from real advertising. It also does the math behind A/B tests.

It's free and needs no API key. It makes no AI calls of its own, either: your model does the thinking, and the server only serves the catalog and runs plain code.

Connect it

The hosted server lives at https://taxonomy.coppica.com/mcp. Any client that accepts a remote MCP server (Streamable HTTP) can use that URL as it is.

Claude Code

Claude Code plugin. This adds the server along with a short skill that tells Claude when to reach for it.

/plugin marketplace add Otha-Labs/persuasion-mcp/plugin install persuasion-taxonomy@persuasion-taxonomy

/plugin marketplace add Otha-Labs/persuasion-mcp/plugin install persuasion-taxonomy@persuasion-taxonomy

Claude, ChatGPT and other apps with custom connectors. Add a custom connector and paste in the URL above. There's nothing to sign in to.

Cursor (.cursor/mcp.json)

VS Code (.vscode/mcp.json)

json

{ "servers": { "persuasion-taxonomy": { "type": "http", "url": "https://taxonomy.coppica.com/mcp" } } }

{ "servers": { "persuasion-taxonomy": { "type": "http", "url": "https://taxonomy.coppica.com/mcp" } } }

Run it on your own machine (any client that launches local servers)

json

{ "mcpServers": { "persuasion-taxonomy": { "command": "npx", "args": ["-y", "@coppica/persuasion-mcp"] } } }

{ "mcpServers": { "persuasion-taxonomy": { "command": "npx", "args": ["-y", "@coppica/persuasion-mcp"] } } }

What your AI can do with it

You don't need to name the tools. Ask for what you want and a capable model reaches for the right one, and in testing it did so in all 16 realistic requests we tried, without being told the server existed.

When someone asks...	The tool that answers
"Write me a landing page / ad / email / sales letter"	`plan_marketing_copy` plans it around the nine questions before a word is written
"Is this good?", "Roast this", "What's wrong with this copy?"	`diagnose_marketing_copy` shows which questions the copy answers, which it skips, and what reads as machine-written
"My ad gets clicks but no sales"	`explain_why_not_converting` works back from the symptom to the question going unanswered
"Give me other ways to build trust", "Make this less generic"	`find_persuasion_techniques` searches the catalog for a less expected move
"What is this technique called?", "How does it work?"	`get_persuasion_technique` returns the full entry, with real examples and when it backfires
"Which of these headlines is best?"	`check_headlines` says what each line is betting on and which two to test
"Is this claim believable?", "Is this button any good?"	`check_marketing_claims` catches vague superlatives, unproven promises and stock buttons
"How long should I run this test?"	`plan_ab_test` sizes the test and frames it so the result teaches you something
"Did my test win?"	`read_ab_test_result` reads the result and warns you when it might be lying

It also offers four ready-made prompts (/roast, /brief, /why-not-converting and /next-test) and the nine questions as a resource you can read in full.

What it keeps

Nothing you send it. Your copy, your brief and your test numbers are used to answer that one request and then dropped. The hosted server logs which tool was called and when, and like any website, its host keeps ordinary request logs for a short time. The full policy is at taxonomy.coppica.com/privacy.

Where the techniques come from

Every technique is an entry in The Persuasion Taxonomy, a catalog of persuasion techniques documented from real ads, sales letters, emails and pages, with examples from the 1920s to now. The catalog is free under CC BY 4.0. When your AI uses a technique, it names the technique and links to its page, so your reader can see the real examples behind the advice.

Develop

bash

npm installnpm run buildnpm run smoke            # connects as a client over stdio and calls every toolnpm run smoke -- --http  # the same, over Streamable HTTPnpm run voice            # checks every model-facing string for machine tells

npm installnpm run buildnpm run smoke            # connects as a client over stdio and calls every toolnpm run smoke -- --http  # the same, over Streamable HTTPnpm run voice            # checks every model-facing string for machine tells

The catalog in data/ is a snapshot. npm run export-data refreshes it, and it needs read access to Coppica's database, so outside contributors can work against the snapshot as it is.

The src/http.ts module exports handleMcpRequest(request), which takes a web-standard Request and returns a Response. That makes it easy to host anywhere that speaks fetch: a Next.js route, a Vercel or Cloudflare function, or plain Node 18 and up.

Voice

Every word the server shows a model reads the way a good direct-response writer would write it: plain words, full sentences that lead into each other, no em dashes, no shouted labels and no "it's not X, it's Y". That goes for the instructions, the tool descriptions, every line of output and every error message. A model picks up the voice of whatever it reads and passes it on to whoever it's writing for, so the tool has to model the writing it asks for. src/voice.ts turns the scorers' findings into plain sentences, and npm run voice keeps it honest.

License

The code is MIT. The catalog data in data/ is CC BY 4.0 and belongs to The Persuasion Taxonomy by Coppica.

About

Built by Coppica. This server knows how readers think. Coppica connects that same model of the reader to your live conversion data, so over time it learns which answers actually sell to your buyers.

來源：README.md，提交 a1dd265

工具

工具後設資料尚未被收錄。

版本歷史

v1.0.0最新Oct 4, 2026