OpenAPI MCP Studio Demo
An MCP server that demonstrates converting a safe OpenAPI subset into MCP tools and interactive apps, using the GitHub API as an example. It provides three GitHub tools with summary, table, and ranking views.
README
OpenAPI-to-MCP Studio Demo
A clean-room product proof for configuring a safe OpenAPI subset as MCP tools and interactive MCP Apps. It uses mcp-use v2 and Manufact Cloud rather than rebuilding MCP infrastructure.
Current Scope
The Studio imports the bundled GitHub Repository API subset, classifies supported operations, and locally edits a bounded app-config.json. The runtime registers the resulting checked-in configuration as three safe GitHub tools with interactive summary, table, and ranking views.
Getting Started
This project requires Node >=22.22.2. On this machine, use the Homebrew Node 24 runtime:
export PATH="/opt/homebrew/opt/node/bin:$PATH"
npm install
npm run dev
Open http://localhost:3000/mcp/inspector, then invoke open-studio to browse the catalog.
The Studio saves edits to browser local storage and can export app-config.json. This is intentionally prototype-only: exported changes do not modify the running server until they are reviewed, copied into data/app-config.json, and the server is restarted.
The three configured GitHub tools are also available locally:
npx mcp-use client connect local http://localhost:3000/mcp
npx mcp-use client local tools call get_repository_overview owner=mcp-use repo=mcp-use
npx mcp-use client local tools call list_repository_issues owner=mcp-use repo=mcp-use state=open limit:=10
npx mcp-use client local tools call list_top_contributors owner=mcp-use repo=mcp-use limit:=10
GITHUB_TOKEN is optional. Copy .env.example to .env.local and set it only when GitHub's unauthenticated rate limit becomes restrictive.
Quality Checks
npm run test
npm run typecheck
npm run build
node .agents/skills/mcp-apps-builder/scripts/check-v2.mjs .
Project Layout
data/app-config.json: checked-in example configuration.examples/: reduced GitHub OpenAPI document and unsupported fixture.src/config/: Zod configuration schema.src/openapi/: OpenAPI parsing and support classification.views/studio/: catalog MCP App view.views/summary-card/,views/data-table/,views/ranked-list/: normalized GitHub result views.src/tests/: configuration and parser tests.
Deployment
The completed Studio is deployed at:
- MCP URL:
https://keen-forge-ocsbv.run.mcp-use.com/mcp - Chat URL:
https://keen-forge-ocsbv.run.mcp-use.com/chat
All three configured GitHub tools and their assigned MCP App views have been verified against this public endpoint.
Security Boundary
- The prototype will support only explicit allowlisted HTTPS hosts at runtime.
- Optional bearer credentials remain server-side and must never be committed.
- Requests are GET-only, do not accept caller-provided headers, reject redirects, enforce a timeout and 1 MB response cap, and log only safe request metadata.
- Unsupported OpenAPI constructs are classified visibly instead of being generated.
- The project contains only public GitHub API examples and clean-room code.
Current Limitations
- The Studio currently starts from the bundled spec; remote URL import is a future enhancement.
- Browser persistence and JSON export are not cloud persistence or collaborative configuration.
- Only the bundled GET operations are registered at runtime; editing the browser config does not hot-reload server tools.
- The current Manufact deployment contains the unmodified Milestone 0 template and will be updated only after final review.
Phase 2 Direction
The next validated product slice is a separately hosted Studio with remote spec import, visual view bindings and preview, versioned configuration, and a constrained flow builder for sequential tools, field mapping, conditions, and explicit user confirmations. See docs/phase-2.md.
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。