PinterestMCP
Read-only MCP server for exposing Pinterest boards, sections, pins, and imagery as a visual reference library. Enables ChatGPT and other clients to list, search, and retrieve Pinterest content through natural language.
README
PinterestMCP
Read-only Pinterest MCP server for ChatGPT and other MCP clients. The goal is to expose Pinterest boards, sections, pins, and pin imagery as a stable visual-reference library.
Status
This repository now contains a working TypeScript MCP server with:
- Streamable HTTP MCP endpoint at
/mcp - HTTP endpoints at
/,/health,/privacy,/terms - Typed Pinterest API client with token refresh support
- Read-only MCP tools for boards, sections, pins, search, and pin images
- Unit and integration tests
- GitHub Actions CI
- Vercel-ready entrypoint and rewrites
Current auth modes:
none: local development and inspector workbearer: simple shared-secret gate for a private endpoint
Important:
- Official OpenAI docs recommend OAuth for user-specific remote MCP data.
- This repo is auth-ready at the application boundary, but a production OAuth authorization server is still a separate step.
- Pinterest runtime verification against the real account requires valid Pinterest credentials in environment variables.
Architecture
flowchart TD
A["ChatGPT or MCP Client"] -->|Streamable HTTP| B["PinterestMCP"]
B -->|Bearer access token or refresh flow| C["Pinterest API v5"]
C --> D["Pinterest Account Data"]
E["Optional MCP Auth Layer"] --> B
Auth layers:
- Layer A:
PinterestMCP -> Pinterest API - Layer B:
ChatGPT -> PinterestMCP
Available MCP Tools
pinterest_list_boardspinterest_get_boardpinterest_list_board_sectionspinterest_list_board_pinspinterest_get_pinpinterest_search_saved_pinspinterest_get_pin_imagepinterest_get_reference_set
All tools are annotated as read-only.
Local Setup
Requirements:
- Node.js 22+
- npm 10+
Install and run:
npm ci
cp .env.example .env
npm run dev
Health check:
curl http://localhost:3000/health
Environment Variables
Minimum local Pinterest setup:
PINTEREST_ACCESS_TOKEN=
Refresh-capable setup:
PINTEREST_APP_ID=
PINTEREST_APP_SECRET=
PINTEREST_REFRESH_TOKEN=
PINTEREST_ACCESS_TOKEN=
PINTEREST_ACCESS_TOKEN_EXPIRES_AT=
MCP gate:
MCP_AUTH_MODE=none|bearer
MCP_BEARER_TOKEN=
Pinterest Developer Setup
Based on the official Pinterest docs checked on August 13, 2026:
- Pinterest API uses
v5. - OAuth token endpoint is
https://api.pinterest.com/v5/oauth/token. - Authorization Code flow is the recommended full-capability flow.
- Pinterest now uses continuous refresh tokens with 60-day expiry and indefinite refresh as long as they are rotated in time.
- Relevant scopes are confirmed in the official docs:
boards:readpins:readuser_accounts:readboards:read_secretpins:read_secret
- Official user pin search exists at
GET /search/pins, so this server uses that endpoint when possible.
OAuth and MCP Auth
Pinterest OAuth and MCP auth are separate:
- Pinterest OAuth protects access to Pinterest data.
- MCP auth protects the PinterestMCP endpoint from arbitrary clients.
OpenAI's current docs state that user-specific remote MCP data should be protected with OAuth rather than an anonymous public endpoint. This repo therefore keeps MCP auth configurable and isolates it from the Pinterest client layer.
Deployment
Vercel is supported through:
Build locally:
npm run build
ChatGPT Connection
Two realistic paths exist:
- Manual MCP connection for local/private testing.
- Workspace app / remote MCP rollout after production auth is added.
Useful checks before connecting:
curl http://localhost:3000/health
curl -X POST http://localhost:3000/mcp -H 'Content-Type: application/json' -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"curl","version":"1.0.0"}}}'
Testing
Run the full local verification suite:
npm run lint
npm run typecheck
npm test
npm run build
What is covered:
- normalizers
- pagination
- search ranking
- error mapping
- image host validation
- MCP initialize + tools/list
- MCP
pinterest_get_pin_imageimage-content delivery
Optional smoke coverage:
tests/pinterest-smoke.test.tsactivates only when Pinterest credentials are present.
Security
See SECURITY.md.
Limitations
- No write operations by design.
- Production-grade MCP OAuth is not implemented yet.
- Real Pinterest verification depends on valid Pinterest credentials.
- Final ChatGPT workspace-app verification depends on OpenAI workspace permissions and browser-side setup.
Troubleshooting
401 unauthorizedon/mcp: checkMCP_AUTH_MODEandMCP_BEARER_TOKEN.pinterest_not_configured: add a Pinterest access token or refresh-capable credentials.image_url_untrusted_host: the pin image host is outside the Pinterest CDN allowlist.- Empty search results with local fallback: provide
boardIdor enable official search.
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。