Supermemory MCP
An MCP server that enables users to visualize Supermemory document data as an interactive memory graph within MCP-compatible clients like Claude. It provides a dedicated tool to fetch and render data as an interactive widget while keeping API keys secure on the server side.
README
Supermemory MCP
Built with Codex. Review, test, and harden before using in production.
MCP server for rendering your Supermemory data as an interactive memory graph inside MCP-compatible clients (Claude, Inspector, and other MCP Apps hosts).
Demo Preview

What This Server Does
- Exposes a tool:
show-memory-graph - Fetches document data from Supermemory on the server side
- Returns an interactive widget powered by
@supermemory/memory-graph - Keeps your Supermemory API key off the client
Architecture
- Runtime:
mcp-useserver - Tool handler:
/index.ts - Widget UI:
/resources/memory-graph/widget.tsx - Widget props schema:
/resources/memory-graph/types.ts - Shared styles:
/resources/styles.css
Prerequisites
- Node.js 18+
- npm
- Supermemory API key
Environment Variables
Required:
SUPERMEMORY_API_KEY=sm_...
Optional:
# Base URL for MCP server metadata (important in deployed environments)
MCP_URL=https://your-mcp-server.example.com
# Override Supermemory API host if needed
SUPERMEMORY_API_BASE_URL=https://api.supermemory.ai
Local Development
npm install
npm run dev
Open Inspector:
http://localhost:3000/inspector
Tool Contract
Tool Name
show-memory-graph
Inputs
page?: numberdefault1limit?: numberdefault250(max500)sort?: "createdAt" | "updatedAt"default"createdAt"order?: "asc" | "desc"default"desc"variant?: "console" | "consumer"default"consumer"showSpacesSelector?: booleandefault depends on variant
Behavior
- Calls
POST /v3/documents/documentson Supermemory - Returns widget props (
documents,variant, etc.) - Returns text output summary for the model
- Handles API errors and surfaces readable error messages in the widget
Deploy
mcp-use deploys source from GitHub, but environment variables are attached to the deployment (not committed to the repo).
npm run deploy
Before deploy:
- Set
SUPERMEMORY_API_KEYin the deployment environment - Set
MCP_URLto the final public URL - Redeploy/restart after environment changes
Deploy with environment variables (recommended)
Set env vars at deploy time:
npm run deploy -- \
--env SUPERMEMORY_API_KEY=sm_your_key_here \
--env SUPERMEMORY_API_BASE_URL=https://api.supermemory.ai
Or use an env file:
# .env.production (do not commit this file)
SUPERMEMORY_API_KEY=sm_your_key_here
SUPERMEMORY_API_BASE_URL=https://api.supermemory.ai
npm run deploy -- --env-file .env.production
If your deployment already exists, re-running deploy updates code from GitHub and keeps/updates deployment env vars based on your flags/dashboard settings.
Using with Claude
- Deploy this MCP server
- Add the MCP server URL in Claude MCP settings
- Call the tool:
show-memory-graph
Important:
- Do not put
SUPERMEMORY_API_KEYinto Claude connector fields - The key belongs only in the MCP server environment
Troubleshooting
401 Unauthorized from Supermemory
- Key is invalid, malformed, or from the wrong environment
- Ensure key value does not include
Bearer - Restart/redeploy after updating env vars
Graph header renders but canvas is blank
- Usually caused by malformed/empty memory payloads or sizing
- This project normalizes memory entries and enforces widget viewport height
- Confirm tool response includes non-empty
documents
Build command fails with mcp-use: command not found
- Dependencies are not installed in the current environment
- Run
npm installfirst
Security Notes
- API key is only used server-side
- Never expose Supermemory credentials in widget/client code
- Review CSP and external domains before production rollout
Production Checklist
- Validate input limits and rate controls
- Add request/response observability and error tracking
- Add auth and access controls for your MCP endpoint
- Load test with real document volume
- Pin dependency versions instead of
latestwhere required
Useful Links
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。