
Kit.com MCP Server
io.github.dancumberland由 GitHub 帳號 dancumberland 發布v1.0.1更新於 Oct 24, 2026
Agent-optimized MCP server for Kit.com email marketing. 13 composite tools, 100% V4 API.
概覽
本機 MCP 伺服器,把助理連接到 Kit.com 電子郵件行銷帳戶,處理訂閱者、標籤、廣播與互動數據。
- 可以做什麼
- 尋找、列出、建立、更新與取消訂閱訂閱者,檢視其統計並比較最多 100 人的互動情況
- 列出、建立與更新標籤,為訂閱者加上或移除標籤,並列出已標記的訂閱者
- 列出、取得、建立草稿、更新與刪除廣播,並檢視廣播統計、跨廣播比較與單一連結點擊資料
- 列出表單、序列、區隔、郵件範本與 webhook,並把訂閱者加入表單或序列
- 建立與刪除自訂欄位和 webhook,取得含成長數據的帳戶總覽
- 執行訂閱者、標籤、表單與欄位的批次操作,並測試 API 連線
- 適用情境
- 適合管理 Kit.com(原 ConvertKit)郵件名單的人:用自然語言詢問名單成長、訂閱者互動、廣播成效與連結點擊情況,或完成加標籤、建立廣播草稿等日常操作,不必開啟 Kit 後台。
- 執行需求
- 透過 npx 在本機執行,需要 Node.js 22 或更新版本。需要 Kit.com 帳戶(免費或付費)以及透過 KIT_API_KEY 設定的 Kit API 金鑰;購買與批次操作還需要選用的 KIT_OAUTH_TOKEN。需要連線至 Kit V4 API 的網路存取。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Kit.com MCP Server,將其新增到工作區。
- 為需要使用其工具的對話啟用該服務。
Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。
其他 MCP 客戶端
參照 儲存庫 中的啟動說明。
README
@dancumberland/kit-mcp
The most complete MCP server for Kit.com (formerly ConvertKit). 13 agent-optimized tools covering 100% of the Kit V4 API — including engagement analytics, bulk operations, and broadcast click tracking that no other Kit MCP offers.
How This Is Different
Other Kit MCP servers wrap each API endpoint as a separate tool (29+ tools). That approach breaks in practice:
Engagement Analytics (Exclusive)
This is the only Kit MCP that can answer "who are my most engaged subscribers?":
Other Kit MCPs can list subscribers and get basic profiles, but can't fetch engagement stats, compare across subscribers, or filter by engagement metrics.
What This Does
Connects any MCP client (Claude Desktop, Claude Code, Cursor, Windsurf, etc.) to your Kit.com email marketing account. Ask questions naturally:
- "How many subscribers do I have and how's my list growing?"
- "Show me my broadcast stats from last week"
- "Find my most engaged subscribers from the past 6 months"
- "Tag everyone who signed up through my landing page"
- "Create a draft broadcast for my newsletter"
- "Which links got the most clicks in my last broadcast?"
No coding required — just set it up and start talking.
Prerequisites
- Node.js 22+ — Download from nodejs.org. Check with
node --version. - A Kit.com account — Free or paid, any plan.
- Your Kit API key — kit.com → Account Settings → Developer. Starts with
kit_.
Setup: Claude Desktop App (Recommended)
Step 1: Find your config file
Mac:
Open claude_desktop_config.json in any text editor.
Windows:
Step 2: Add the Kit MCP server
If the file is empty or doesn't exist, paste this (replace your-kit-api-key with your actual key):
If you already have other MCP servers, add "kit" inside the existing "mcpServers" block with a comma after the previous entry.
Step 3: Restart Claude Desktop
Fully quit (not just close the window) and reopen.
Step 4: Verify it works
Test my Kit connection
Claude should respond with your account name, auth method, and rate limit.
Setup: Claude Desktop with Cowork
Cowork is Claude Desktop's background agent. After completing the setup above, Cowork automatically has access to your Kit tools.
Example tasks:
- "Every morning at 8am, summarize my subscriber growth and email performance from the last 24 hours."
- "Every Monday at 9am, compare my broadcast stats from the past week — open rates, click rates, and unsubscribes."
- "Every Friday, list all tags with fewer than 10 subscribers (candidates for cleanup)."
Setup: Claude Code (CLI)
Add to .claude/settings.local.json or ~/.claude/settings.json:
Setup: Cursor
- Open Settings (Cmd+,)
- Search for "MCP"
- Click "Add MCP Server"
- Add:
Tools
Things You Can Ask
Authentication
API Key (covers most features): Get at kit.com → Developer. Set as KIT_API_KEY.
OAuth Token (optional, for purchases and bulk operations): Set KIT_OAUTH_TOKEN alongside your API key.
Rate limits enforced automatically: 120 req/min (API key) or 600 req/min (OAuth), with retry + backoff on 429s.
Architecture: Why 13 Tools Instead of 29+
Most MCP servers create one tool per API endpoint. For Kit's API, that means 29+ tools — which causes real problems:
- Context bloat — 8,000+ tokens just for tool definitions, leaving less room for your actual conversation
- Poor accuracy — AI tool selection degrades measurably beyond 20 tools (research)
- Compatibility — Cursor has a hard limit of 40 tools across all servers; 29 tools from one server leaves almost no room for others
This server uses 13 composite tools with a discriminated action parameter. Same API coverage, 60% fewer tokens, better accuracy. Each tool groups related operations (e.g., all subscriber actions under manage_subscribers) so the AI picks the right tool on the first try.
Error Handling
Errors include recovery hints that Claude can act on:
- 429 (rate limit): Automatic retry with exponential backoff, up to 3 attempts
- 5xx (server error): Automatic retry once
- 422 (validation): No retry — returns the error immediately with a fix suggestion
Troubleshooting
Development
License
MIT
來源:README.md,提交 e2bfa8e
工具
0版本歷史
1- v1.0.1最新Sep 16, 2026
評分
已滿足適用檢查項的 50%
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25 分
請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。
- 已宣告認證方式10 分
請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。
- 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