MCP Toolkit
A free MCP server that adds read-only tools for current time, precise math, and text stats, plugging into both Claude and ChatGPT.
README
MCP Toolkit 🧰
A tiny, free, open-source plugin for both Claude and ChatGPT.
It's a single MCP (Model Context Protocol) server. Because Claude and ChatGPT both now speak MCP, the same server plugs into both — you build it once and one public URL works everywhere.
Tools it adds to the AI
| Tool | What it does | Why it helps |
|---|---|---|
get_current_time |
Real current time in any timezone | LLMs don't actually know "now" |
calculate |
Precise arithmetic (+ - * / % ^, parentheses) |
LLMs make math mistakes |
word_count |
Exact word / character / sentence stats | LLMs can't reliably count |
All three are read-only — they can't change or delete anything.
The one thing to understand first
GitHub stores your code for free, but it can't run a server. (GitHub Pages only serves static files.) To be added to live Claude/ChatGPT, your server needs a public
https://URL that's actually running. So we host the code on GitHub and run it on Cloudflare Workers' free tier.
Your code on GitHub ──deploy──▶ Cloudflare Workers (free) ──▶ https://mcp-toolkit.<you>.workers.dev/mcp
(free) (runs 24/7) │
├──▶ paste into Claude
└──▶ paste into ChatGPT
Cloudflare Workers' free tier is 100,000 requests/day — plenty for personal use, and it never sleeps.
What you need (all free)
- Node.js 18+ installed
- A free Cloudflare account
- A free GitHub account (to publish the code)
Step 1 — Get the code & run it locally
# from the mcp-toolkit folder
npm install
npm run dev
You'll see it running at http://localhost:8787. Open http://localhost:8787/ in a browser — you should
see a "✔ live" page. (Your local MCP endpoint is http://localhost:8787/mcp.)
Tip: to sanity-check the code before deploying, run
npm run type-check.
Step 2 — Deploy it (get your public URL)
npx wrangler deploy
The first time, Wrangler opens a browser to log into Cloudflare. When it finishes it prints your live URL:
https://mcp-toolkit.<your-account>.workers.dev
Your connector URL is that address with /mcp on the end:
https://mcp-toolkit.<your-account>.workers.dev/mcp
Open the base URL in a browser to confirm you see the "✔ live" page. Keep the /mcp URL handy — you'll
paste it into Claude and ChatGPT next.
Step 3 — Put the code on GitHub (make it public & free)
git init
git add .
git commit -m "MCP Toolkit: free MCP plugin for Claude & ChatGPT"
git branch -M main
git remote add origin https://github.com/<your-username>/mcp-toolkit.git
git push -u origin main
That's it — the code is now public and anyone can use it for free (MIT license).
Optional — auto-deploy on every push: this repo includes .github/workflows/deploy.yml.
Add a CLOUDFLARE_API_TOKEN secret in your repo (Settings → Secrets and variables → Actions), and every
git push will redeploy automatically. (Details are in the workflow file's comments.)
Step 4 — Add it to Claude
Works on Free, Pro, Max, Team, and Enterprise (Free is limited to one custom connector).
- Open claude.ai → your profile/Settings → Connectors (it may be under Customize → Connectors).
- Click Add → Add custom connector.
- Paste your URL:
https://mcp-toolkit.<your-account>.workers.dev/mcp - Click Add. (Leave the OAuth fields blank — this server needs no login.)
- In any chat, click the + menu → Connectors, enable MCP Toolkit, then ask: "What time is it right now in Asia/Kolkata?" or "Use calculate: (349 * 12.5) / 100".
Works the same in Claude Desktop (Settings → Connectors → Add custom connector). On Team/Enterprise, only an organization Owner can add it.
Step 5 — Add it to ChatGPT
ChatGPT reaches custom MCP tools through Developer Mode (a beta). Requirements: Plus, Pro, Business, Enterprise, or Edu, on ChatGPT in a desktop web browser (not the mobile app).
- ChatGPT (desktop web) → Settings → Connectors → Advanced → turn on Developer mode. (On Business/Enterprise, an admin may need to enable this first.)
- Back in Connectors, click Create / Add. Fill in:
- Name:
MCP Toolkit - Description:
Current time, precise math, and text stats - URL:
https://mcp-toolkit.<your-account>.workers.dev/mcp - Authentication: None
- Name:
- Save. In a chat, open the + / tools menu, enable MCP Toolkit, and ask the same kind of question.
Why Developer Mode? ChatGPT's older "Deep Research connectors" only accept servers that expose exactly two tools (
search+fetch). Developer Mode lifts that restriction so any MCP tools work in normal chat. If your plan can't use Developer Mode, this general-purpose toolkit won't fit the Deep-Research path — but it works fully in Claude.
Add your own tool (extend it)
Open src/index.ts and copy the pattern inside init():
this.server.registerTool(
"reverse_text",
{
title: "Reverse text",
description: "Reverse a string of text.",
inputSchema: { text: z.string() }, // a plain object of zod fields
annotations: { readOnlyHint: true }, // read-only = no confirmation prompt
},
async ({ text }) => ({
content: [{ type: "text", text: [...text].reverse().join("") }],
}),
);
Then npx wrangler deploy again (or just git push if you set up auto-deploy). Reconnect in Claude/ChatGPT
to pick up the new tool.
Troubleshooting
| Problem | Fix |
|---|---|
| Connector won't connect | Make sure you used the /mcp URL and that opening the base URL in a browser shows the "✔ live" page. The server must be publicly reachable (Cloudflare's is). |
npx wrangler deploy fails to log in |
Run npx wrangler login once, then retry. |
| ChatGPT has no "Developer mode" | It needs a Plus+ plan on desktop web; on Business/Enterprise an admin must enable it. |
| Tools don't appear in a chat | Open the + / connectors menu in that chat and toggle the connector on. |
| Don't put secrets in the URL | Never add ?token=... to the URL — it's blocked by the MCP spec. Use proper auth if you need it. |
How it's built (for the curious)
- Runtime: Cloudflare Workers (serverless, free tier, never sleeps)
- Library:
agents'McpAgent+ the official@modelcontextprotocol/sdk - Transports served:
/mcp(Streamable HTTP, recommended) and/sse(legacy, for older clients) - Auth: none (authless) — anyone with the URL can use the read-only tools
License
MIT — free for anyone to use, modify, and share. Replace <Your Name> in LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。