Mora AI

io.github.Mora-AI-Content-Studio由 GitHub 帳號 Mora-AI-Content-Studio 發布v1.0.3更新於 Oct 24, 2026

Codex guesses your brand voice. Mora AI gives it the real one: your catalogue and performance.

已驗證Streamable HTTP可網頁執行效率與工作流資料與分析商業與電商

概覽

AI 產生的概覽

Mora AI 是遠端唯讀 MCP 連接器,讓助理取得創作者真實的品牌資料、商品目錄、貼文成效與廣告數據,用於有依據的內容規劃。

可以做什麼
  • 讀取貼文及其成效,以及 Google、Meta 和 LinkedIn Ads 的廣告投放資料
  • 讀取品牌資料、簡報工作區、專案、受眾與內容角度
  • 讀取商品目錄以及 Shopify 營收歸因
  • 每個工作階段可掛載一次品牌語調資源,讓草稿維持品牌一致
  • 執行發布流程提示詞,例如 Product Hunt 發布週、首則創作者留言、依既有成功內容寫作
  • 使用內建技能做成效稽核、品牌簡報、內容缺口檢查、付費廣告文案與行銷活動計畫
適用情境
適合已在使用 Mora 的創作者、教練與小型團隊:在助理工作階段中規劃發布、起草貼文或廣告文案,或查看哪些內容已經奏效,並依據自己帳戶的真實資料而非猜測的品牌語調。
執行需求
遠端端點位於 app.mora-marketer.com,不需本機套件或 API 金鑰。需要既有或新建立的 Mora 帳戶,每個請求都需要透過 Mora 同意頁面取得的 OAuth 持有者權杖。伺服器使用 CIMD 用戶端識別,README 稱 Claude Code 受支援,Claude Desktop 預期可用但未經獨立驗證,Codex CLI、Cursor、VS Code 與 Windsurf 尚不支援。
安裝前請注意
存取為唯讀:任何工具都不能發布、排程、連接帳戶、變更廣告帳戶或花錢。廣告工具需要初始 mora.read 之外的個別 mora.read.ads 授權,帳戶擁有者會被要求核准廣告報表。品牌資料、商品目錄、成效與營收歸因等帳戶資料會傳送到所連接的助理工作階段。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Mora AI,將其新增到工作區。
  2. 連接 所需的 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
  3. 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。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/io.github.Mora-AI-Content-Studio%2Fmora/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。 客戶端會在首次連接時完成 OAuth 登入。

{
  "mcpServers": {
    "mora-ai": {
      "type": "http",
      "url": "https://app.mora-marketer.com/api/mcp"
    }
  }
}

README

Mora AI

Codex guesses your brand voice. Mora AI gives it the real one, plus your real catalogue and performance.

What is Mora AI

Mora serves solo creators and small teams building an audience or a brand, not only stores. That covers a Shopify or DTC operator who wants a product named correctly instead of invented, a coach or speaker building a personal brand on LinkedIn with no catalogue at all, and a full-time creator posting daily across three platforms. Mora also has a franchise tier for multi-location owners and an enterprise tier for organizations running governed brand templates. It plans, drafts, gets human approval, schedules and publishes content, and closes the loop with real engagement and, for connected Shopify stores, revenue attribution. What Mora does that a generic content tool doesn't: it grounds every draft in the account's own real data, not a brand-voice guess.

Who this plugin is for: a user of a supported MCP host who already has (or wants) a Mora account and wants to pull their own brand context, past performance, and product catalogue into an agent session — for example, to plan a Product Hunt launch, draft a week of posts, or check what already worked before writing more.

What this plugin gives you

  • Mora AI's official remote MCP server connector, pre-configured — no manual .mcp.json editing, no API key to paste. First use triggers Mora's own OAuth 2.1 consent screen in your browser. The hosted MCP runtime and account-scoped handlers live in Mora's product app; this public repository contains only connector configuration and workflow skills.
  • Read-only tools: list_posts, get_post_performance, get_brand_profile, get_brief_workspace, list_products, get_revenue_attribution, list_projects, list_audiences, list_content_angles, list_brief_runs, and get_paid_campaign_performance — Google, Meta, and LinkedIn Ads campaign delivery as each platform reports it. That last tool needs a separate mora.read.ads consent: a new connection asks only for mora.read, and the first time an agent calls the ads tool, Mora asks the account owner to approve ad reporting on its own consent line. No tool can publish, schedule, connect an account, change an ad account, or spend money — Mora treats writes through an unattended agent loop as unsafe to expose.
  • A brand-voice resource (mora://brand/voice) you can attach once per session so every message Claude writes stays on-brand, without re-fetching it per message.
  • Three launch-workflow prompts — product_hunt_launch_week, first_maker_comment, and write_like_what_worked — most notably a full Product Hunt launch week that already knows Product Hunt's current rules (no vote solicitation, no "Coming Soon" pages — retired August 2025) so you don't have to.
  • Six skills that package the tools above into actual workflows instead of leaving you to chain tool calls yourself:
    • skills/mora-mcp-setup — first connection, what each tool and prompt is for. Triggers when you ask about connecting to or using Mora.
    • skills/content-performance-audit — combines list_posts and get_post_performance into a synthesized brief on what's actually working (platform, theme, timing) and what to do next. Triggers on "how is my content performing", "audit my Mora posts", "what's working on my social".
    • skills/brand-brief — combines get_brand_profile and list_products into a portable, factual brand + catalogue summary for work happening outside Mora — briefing a designer, writing ad copy, prepping a pitch. Triggers on "give me my Mora brand brief", "what does my brand sell", "summarize my Mora catalogue".
    • skills/content-gap-check — combines list_projects and list_posts to compare what's planned against what's actually gone out and surface overdue or orphaned content. Triggers on "check my Mora content calendar for gaps", "am I behind on my content schedule".
    • skills/paid-ad-copy — turns real brand, catalogue, audience, angle, and performance data into paid-ad copy and a human-reviewable test matrix. It never creates campaigns or spends money.
    • skills/campaign-plan — turns existing plans, posts, audiences, angles, catalogue facts, and performance into a grounded campaign plan without scheduling or publishing anything.

Why read-only, and why so few tools

Read-only is a considered position, not a placeholder. Every serious content-quality failure Mora has shipped and documented — a fabricated brand wordmark reaching a draft, unrequested posts silently added to an account — happened with a human reviewing the loop. Handing write access to a loop without one isn't something Mora can honestly call safe yet.

A small tool count is a measured ceiling, not an economy move: published research on tool-selection accuracy shows smaller models' tool-choice accuracy degrading past roughly 10–15 tools in a single context, and Mora's own design principle is that a longer tool list is not automatically better for an agent holding it. These answer the questions a marketing agent actually asks — including "what real revenue does a channel actually deserve credit for", "which audience is this for", and "which content angles or brief findings already exist" — and nothing here overlaps.

Requirements

  • An existing or new Mora account (free to create at https://www.mora-marketer.com).
  • Claude Code with plugin support and remote MCP + OAuth support enabled.

Contract and runtime ownership

The canonical remote MCP endpoint is https://app.mora-marketer.com/api/mcp. The live server card at https://app.mora-marketer.com/.well-known/mcp/server-card.json is the source of truth for the tool, resource, prompt, and read-only metadata. This public repository is the Claude Code connector and workflow package; it is not the private MCP runtime. Repository: https://github.com/Mora-AI-Content-Studio/mora-claude-plugin.

Run the production contract check before publishing documentation changes:

bash
node scripts/check-mora-mcp-contract.mjs

For an offline fixture, set MORA_MCP_SERVER_CARD_FILE=/path/to/server-card.json.

Developer documentation:

Authentication and client compatibility

Every request needs an OAuth-issued bearer token. Mora's server identifies clients with CIMD (Client ID Metadata Documents) — the MCP spec's current recommended standard (2026-07-28), which formally deprecates the older Dynamic Client Registration (DCR) flow. Claude's own SDKs and Claude Code already ship CIMD; most other MCP clients are still catching up to the new spec, since it's a recent change.

ClientSupport
Claude CodeSupported — this repository is its connector.
Claude DesktopExpected to work (Anthropic's SDK ships CIMD); not yet independently verified.
Codex CLI, Cursor, VS CodeNot yet — these clients' OAuth flow currently expects DCR, which this spec-forward server doesn't implement. Tracking upstream support; see docs/mcp-contract.json's clientCompatibility field for verification dates and test detail.
WindsurfNot yet, for a different reason — Windsurf's remote MCP support is API-key only as of this writing, and this server has no API-key program at all, OAuth only.

This is an authentication-protocol gap, not a missing feature on Mora's side — CIMD is where the ecosystem is heading, not a workaround. docs/mcp-architecture.md has the full auth-flow diagram and the live test evidence behind the table above.

This isn't a rendering gap other clients will "just work around" — until a client ships CIMD support (or Mora separately stands up a DCR-compatible path, which is not currently planned), the honest answer for an unverified/failing client is "not yet," not "add the URL and see." Machine-readable form of this table: docs/mcp-contract.json's clientCompatibility field. See docs/mcp-architecture.md for the full request/auth flow diagram.

Getting started

Ask the connected agent something like "connect to my Mora account" and the bundled skill walks you through the one-time OAuth consent. Then try:

  • "What already worked before I draft this week's posts?"
  • "Plan my Product Hunt launch week, and follow Product Hunt's current rules."
  • "Name a real product from my catalogue instead of guessing one."

Package targets

Claude Code has the documented turnkey installation path above. The same repository also includes .codex-plugin/plugin.json so Mora AI can be packaged and reviewed as a Codex plugin. That manifest does not change the hosted OAuth flow or upgrade any row in the client-compatibility matrix: the current contract records Codex CLI as not yet supported, Cursor and VS Code as unverified, and Windsurf as blocked by its API-key-only remote MCP path.

Support

License

MIT — see LICENSE.

來源:README.md,提交 56195e9

工具

0
該伺服器需要登入,因此 SourceWeft 無法在沒有帳號的情況下讀取其工具,也未收錄。已連接帳號所能看到的內容可能因帳號及其權限而異。

版本歷史

1
  1. v1.0.3最新Sep 16, 2026

評分

已滿足適用檢查項的 85%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分