Capacities MCP Worker
A Cloudflare Worker that wraps the Capacities API as a remote MCP server, enabling Claude on any device to interact with your Capacities knowledge base.
README
Capacities MCP Worker
A Cloudflare Worker that wraps the Capacities API as a remote MCP server. Connect Claude on any device -- iPhone, web, desktop, Claude Code -- to your Capacities knowledge base.
Why this exists
The community capacities-mcp package runs locally (stdio transport).
That works for Claude Code and Claude Desktop, but not for claude.ai (web)
or Claude on iOS/Android, which require a remote MCP server over HTTPS.
This Worker is that remote bridge.
Architecture
Claude (web / iOS / Desktop / Code)
|
| HTTPS (Streamable HTTP or SSE)
v
CF Worker (this project)
|
| HTTPS + Bearer token
v
Capacities API (api.capacities.io)
Setup
1. Prerequisites
- Capacities Pro subscription ($8/mo) with API key
- Cloudflare account (free tier is enough)
- Node.js 18+ and npm
2. Clone and install
cd capacities-mcp-worker
npm install
3. Set your Capacities API key as a secret
npx wrangler secret put CAPACITIES_API_KEY
# paste your key from Capacities Settings > API
4. Deploy
npm run deploy
Wrangler prints your URL, e.g.:
https://capacities-mcp.<you>.workers.dev
5. Test
curl https://capacities-mcp.<you>.workers.dev/health
# {"status":"ok","service":"capacities-mcp"}
6. Connect Claude (web + iOS)
- Go to https://claude.ai
- Settings > Connectors > Add Custom Connector
- Name: "Capacities"
- URL:
https://capacities-mcp.<you>.workers.dev/mcp - Save
The connector syncs to your iPhone Claude app automatically.
7. Connect Claude Desktop (optional, alternative to local capacities-mcp)
Add to your Claude Desktop config:
{
"mcpServers": {
"capacities": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://capacities-mcp.<you>.workers.dev/sse"]
}
}
}
8. Connect Claude Code (optional)
claude mcp add capacities \
--transport sse \
--url https://capacities-mcp.<you>.workers.dev/sse
Or keep using the local capacities-mcp package for Claude Code (faster,
no network hop).
MCP Tools
| Tool | Description |
|---|---|
capacities_list_spaces |
List all your spaces |
capacities_get_space_info |
Get structures, collections, properties for a space |
capacities_search |
Search content by keyword (title or full-text) |
capacities_save_weblink |
Save a URL with tags, description, markdown notes |
capacities_save_to_daily_note |
Append markdown to today's daily note |
capacities_create_object |
Create any object type (page, custom types, etc.) |
Example prompts
- "List my Capacities spaces"
- "Save this to my daily note: learned about Cloudflare R2 today"
- "Create a page called 'Ito's Lemma' with tags quant-finance and math"
- "Search for notes about stochastic calculus"
- "Save this URL to my research space: https://example.com/paper"
Security
- The Capacities API key is stored as a Cloudflare Worker secret (encrypted)
- The Worker does not add its own auth layer (the MCP endpoint is public)
- If you want to restrict access, add Cloudflare Access or a bearer token check in the fetch handler (see comments in index.ts)
To add a simple bearer token gate, set a secret:
npx wrangler secret put MCP_AUTH_KEY
Then uncomment the auth check in src/index.ts.
Limitations
- Capacities API cannot read full object content (search returns IDs/titles only)
- Rate limited to 120 requests per 60s per endpoint
- API is beta and subject to change
Cost
- Capacities Pro: ~$8/mo
- Cloudflare Worker: $0 (free tier: 100k requests/day)
- Durable Objects: $0 for low usage
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。