Pinpole MCP Server
Enables AI agents to design cloud architectures, run cost/performance simulations, and draw them on a Pinpole canvas directly from Claude Code, Cursor, and OpenAI Codex.
README
Pinpole MCP Server
Design cloud architectures, run cost/performance simulations, and draw them on your Pinpole canvas — directly from Claude Code, Cursor, and OpenAI Codex. A draw.io replacement driven by your AI agent.
Product: https://pinpole.cloud · App: https://app.pinpole.cloud
What it does
Your coding agent can now talk to Pinpole over the Model Context Protocol:
| Tool | What it does | Auth |
|---|---|---|
pinpole_build_architecture |
Flagship. Prompt → validated architecture → drawn on your canvas. Returns a canvas URL. Optional cost simulation. | token |
pinpole_create_architecture |
Generate a {nodes, edges} graph from a prompt (no drawing). |
token |
pinpole_simulate_cost |
Run the simulation engine over a graph at a traffic level → per-node + total monthly cost, latency p50/p95/p99, throttling alerts. | none |
pinpole_draw_on_canvas |
Persist a graph to your canvas (creates project/workspace if needed). Auto-layout. | token |
pinpole_open_canvas |
Get the canvas deep link for a project/workspace. | none |
pinpole_list_projects / pinpole_create_project |
Manage projects. | token |
pinpole_export_terraform |
Generate Terraform from a graph (offline). | none |
pinpole_list_services |
List AWS service ids the canvas understands (local repo only). | none |
A typical flow: the agent calls pinpole_build_architecture with a description, Pinpole's
AWS Well-Architected model designs it, the nodes + connections appear on your canvas, and the
agent hands you a URL to open the live, editable diagram.
1. Get a token
Open Pinpole → Settings → Developer / MCP and create a personal access token
(pp_live_…). Copy it — it's shown only once.
Tokens act on behalf of your account (create projects/workspaces, draw on the canvas, use AI-architect credits). Revoke any token from the same screen.
2. Configure your agent
Claude Code
claude mcp add pinpole \
--env PINPOLE_API_TOKEN=pp_live_… \
--env PINPOLE_BASE_URL=https://app.pinpole.cloud \
-- npx -y @pinpole/mcp
Or commit .mcp.json to a repo (see examples/claude-code.mcp.json).
Cursor
Add to ~/.cursor/mcp.json (or .cursor/mcp.json in a project) — see
examples/cursor.mcp.json:
{
"mcpServers": {
"pinpole": {
"command": "npx",
"args": ["-y", "@pinpole/mcp"],
"env": {
"PINPOLE_API_TOKEN": "pp_live_…",
"PINPOLE_BASE_URL": "https://app.pinpole.cloud"
}
}
}
}
OpenAI Codex
Add to ~/.codex/config.toml — see examples/codex-config.toml:
[mcp_servers.pinpole]
command = "npx"
args = ["-y", "@pinpole/mcp"]
env = { PINPOLE_API_TOKEN = "pp_live_…", PINPOLE_BASE_URL = "https://app.pinpole.cloud" }
Environment variables
| Variable | Default | Notes |
|---|---|---|
PINPOLE_API_TOKEN |
– | Personal access token (pp_live_…). Sent as Authorization: Bearer. |
PINPOLE_BASE_URL |
https://app.pinpole.cloud |
Point at a local server for testing. |
PINPOLE_DEV_USER_ID |
– | Local dev only. Sent as x-pinpole-dev-userid (requires the server to run with ALLOW_DEV_USER_HEADER=1). Lets you test without a token. |
Local development & testing
From the Pinpole repo:
npm run mcp:build # compile mcp/ → mcp/dist
ALLOW_DEV_USER_HEADER=1 npm run dev # start the app (default :3031 via `npm start`, :3000 via dev)
Then point your agent at the local build with the dev header — see
examples/local-dev.mcp.json. PINPOLE_DEV_USER_ID is your
Firebase uid.
Smoke-test the tool list without an agent:
printf '%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"x","version":"0"}}}' \
'{"jsonrpc":"2.0","method":"notifications/initialized"}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' \
| node mcp/dist/server.js
Or use the inspector: npx @modelcontextprotocol/inspector node mcp/dist/server.js.
HTTP / remote connector mode
npm run mcp:http serves a Streamable-HTTP endpoint at POST /mcp (with the minimal OAuth
scaffolding Claude's custom-connector probe expects). Env: MCP_HOST, MCP_PORT (default 3333).
Links
- Pinpole — https://pinpole.cloud
- App — https://app.pinpole.cloud
- Issues & source — https://github.com/codeforstartups/pinpole-mcp
MIT licensed.
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。