Cogniz Memory Platform MCP Server
Official Model Context Protocol (MCP) server for Cogniz Memory Platform. Enables AI assistants like Claude to store and retrieve memories across conversations.
README
Cogniz Memory Platform - MCP Server
Official Model Context Protocol (MCP) server for Cogniz Memory Platform - Enables AI assistants like Claude to store and retrieve memories across conversations.
🌐 Live Server: https://app.cogniz.online/mcp
✨ Features
- 🔌 MCP Protocol 2025-03-26 - Latest streamable HTTP transport
- 🔐 Multi-Tenant - Each user uses their own API key
- 🧠 Persistent Memory - Store and retrieve context across sessions
- 🔍 Semantic Search - Find relevant memories using natural language
- 📁 Project Organization - Organize memories by project
- 💾 Auto-Compression - 65% storage savings with lossless compression
- 🌍 Remote Access - Works with Claude Desktop AND Web UIs
🚀 Quick Start
For Claude Desktop Users
1. Get Your API Key:
- Visit cogniz.online/dashboard
- Navigate to API Reference section
- Copy your API key (format:
mp_1_XXXXXXXXXXXX)
2. Configure Claude Desktop:
Open your Claude Desktop config file:
- Windows:
%APPDATA%\Claude\claude_desktop_config.json - macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json
Add this configuration:
{
"mcpServers": {
"cogniz-memory": {
"url": "https://app.cogniz.online/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY_HERE"
}
}
}
}
Replace YOUR_API_KEY_HERE with your actual API key.
3. Restart Claude Desktop
4. Test It:
You: "Store a memory: MCP integration is working!"
Claude: "I've stored that memory in your Cogniz account."
Verify it appears in your dashboard.
For Web UI Users (ChatGPT Web UI, Claude Web UI, etc.)
Web UIs often can't send custom headers, so we support query parameter authentication.
Method 1: Custom Connector (if supported)
Some Web UIs support custom connectors:
- Go to Settings → Connectors
- Add Custom Connector:
- Name: Cogniz Memory Platform
- URL:
https://app.cogniz.online/mcp?api_key=YOUR_API_KEY - Method: POST
Method 2: URL with API Key
Use this URL format:
https://app.cogniz.online/mcp?api_key=YOUR_API_KEY
⚠️ Security Note: Query parameters are less secure than headers because they appear in logs and URLs. Use this method only for Web UIs that don't support Authorization headers.
🛠️ Available Tools
Once connected, Claude can use these MCP tools:
cogniz_store_memory
Store a new memory in the Cogniz platform.
Parameters:
content(required): The text/information to storeproject_id(optional): Project identifier (default: "default")project_name(optional): Human-readable project namecategory(optional): Category tag (e.g., "code-snippets", "meeting-notes")
Example:
"Store this code snippet in my development project: async function fetchData() { ... }"
cogniz_search_memories
Search memories using semantic search.
Parameters:
query(required): Search textproject_id(optional): Limit search to specific projectlimit(optional): Max results (1-100, default: 10)
Example:
"Search my memories for API authentication examples"
cogniz_list_projects
List all your projects.
Example:
"Show me all my projects"
cogniz_get_stats
View your usage statistics.
Returns:
- Current plan
- Memory usage
- API calls
- Project count
- Storage stats
Example:
"How much memory am I using?"
cogniz_delete_memory
Delete a specific memory by ID.
Parameters:
memory_id(required): ID of memory to delete
Example:
"Delete memory mem_12345"
🔐 Authentication Methods
This server supports two authentication methods to work with different clients:
Method 1: Authorization Header (Recommended)
Best for: Claude Desktop, API clients, secure environments
Format:
POST /mcp HTTP/1.1
Authorization: Bearer mp_1_YOUR_API_KEY
Content-Type: application/json
Pros:
- ✅ More secure
- ✅ Not visible in URLs/logs
- ✅ Standard HTTP authentication
Method 2: Query Parameter
Best for: Web UIs that can't send custom headers
Format:
https://app.cogniz.online/mcp?api_key=mp_1_YOUR_API_KEY
Pros:
- ✅ Works with Web UIs
- ✅ No header support needed
Cons:
- ⚠️ Less secure (visible in URLs)
- ⚠️ Appears in server logs
- ⚠️ May be cached by proxies
📊 Pricing Plans
The MCP server is free to use. You only pay for your Cogniz Memory Platform account:
| Plan | Price | Memory Limit | Projects | API Calls/Month |
|---|---|---|---|---|
| Starter | Free (30 days) | 100 MB | 3 | 1,000 |
| Plus | $7/month | Unlimited | 15 | 15,000 |
| Pro | $49/month | Unlimited | Unlimited | 100,000 |
| Enterprise | Custom | Unlimited | Unlimited | Unlimited |
🧪 Testing
Health Check
curl https://cogniz-claude-mcp.onrender.com/health
Expected Response:
{
"status": "healthy",
"service": "cogniz-mcp-server"
}
Test Authentication (Header Method)
curl -X POST https://app.cogniz.online/mcp \
-H "Authorization: Bearer mp_1_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/list"
}'
Test Authentication (Query Method)
curl -X POST "https://app.cogniz.online/mcp?api_key=mp_1_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/list"
}'
Expected: List of 5 available MCP tools
🏗️ Self-Hosting
Want to run your own instance? Clone and deploy:
Prerequisites
- Node.js 18+
- TypeScript
- Render account (or any Node.js hosting)
Deploy to Render
1. Fork this repository
2. Create Web Service on Render:
- Connect your GitHub fork
- Build Command:
npm install && npm run build - Start Command:
npm start - Environment Variables:
COGNIZ_BASE_URL=https://cogniz.onlineCOGNIZ_PROJECT_ID=default- Optional:
COGNIZ_API_KEY(for demo/testing only)
3. Configure Custom Domain (Optional):
- Professional plan required ($19/month)
- Add CNAME:
your-subdomain.com→your-service.onrender.com - SSL auto-configured
4. Update OAuth Discovery URL:
In src/server-remote.ts line 224:
resource: "https://your-domain.com/mcp",
5. Deploy and test!
📚 Documentation
- MCP Protocol: https://modelcontextprotocol.io
- Cogniz Platform: https://cogniz.online/documentation
- API Reference: https://cogniz.online/dashboard#api-reference
- Support: https://cogniz.online/contact
🔧 Development
Local Setup
# Clone repository
git clone https://github.com/cognizonline/Cogniz_Claude_MCP.git
cd Cogniz_Claude_MCP
# Install dependencies
npm install
# Build TypeScript
npm run build
# Set environment variables
export COGNIZ_BASE_URL=https://cogniz.online
export COGNIZ_API_KEY=mp_1_YOUR_TEST_KEY
# Start server
npm start
# Server runs on http://localhost:3000
Test Locally
# Health check
curl http://localhost:3000/health
# List tools
curl -X POST http://localhost:3000/mcp \
-H "Authorization: Bearer mp_1_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
🤝 Contributing
Contributions welcome! Please:
- Fork the repository
- Create a feature branch
- Make your changes
- Test thoroughly
- Submit a pull request
📋 Troubleshooting
"Authentication required" error
Cause: No API key provided
Fix:
- Desktop: Add
Authorization: Bearer YOUR_KEYto config - Web UI: Add
?api_key=YOUR_KEYto URL
"Invalid API key" (401 error)
Cause: API key is wrong or expired
Fix:
- Get fresh API key from dashboard
- Verify format:
mp_1_XXXXXXXXXXXX
"Rate limit exceeded" (429 error)
Cause: Exceeded monthly API call limit
Fix:
- Check usage in dashboard
- Upgrade plan for more calls
- Limit resets monthly
Connection timeout
Cause: Server on free Render plan (cold starts)
Fix:
- Wait 30-60 seconds for warmup
- Or upgrade to Professional plan ($19/month) for always-on
💰 Hosting Costs
Free Render Plan
- ✅ Free forever
- ⚠️ Cold starts (30-60s delay after 15min idle)
- ⚠️ Limited resources
Professional Render Plan ($19/month)
- ✅ Always-on (no cold starts)
- ✅ Better performance
- ✅ Custom domains
- ✅ Priority support
Recommended for production use.
🔒 Security
API Key Safety
- ✅ Keys transmitted via HTTPS only
- ✅ Server doesn't store keys (stateless)
- ✅ Each request isolated
- ✅ SSL/TLS encryption
Best Practices
- Never commit API keys to git
- Use Authorization header when possible (more secure than query params)
- Rotate keys periodically
- Monitor usage for suspicious activity
- Use query params only for Web UIs that don't support headers
🌐 Protocol Details
MCP Version
2025-03-26 (Streamable HTTP)
Transport
HTTP POST with streaming support
Endpoints
POST /mcp- Main MCP endpointGET /health- Health checkGET /.well-known/oauth-protected-resource- Auth discovery
Supported Methods
tools/list- List available toolstools/call- Execute a toolresources/list- List resources (future)
📞 Support
- Issues: GitHub Issues
- Email: support@cogniz.online
- Documentation: cogniz.online/documentation
📄 License
MIT License - See LICENSE file for details
🙏 Acknowledgments
- Built on Model Context Protocol by Anthropic
- Powered by Cogniz Memory Platform
- TypeScript + Express + MCP SDK
🔄 Changelog
v1.2.0 (Latest - January 2025)
- ✅ Multi-tenant support with user-provided API keys
- ✅ Authorization header authentication (recommended)
- ✅ Query parameter authentication (for Web UIs)
- ✅ Improved error messages
- ✅ OAuth discovery metadata
v1.1.0 (October 2024)
- ✅ Updated to MCP Streamable HTTP protocol (2025-03-26)
- ✅ Fixed API field mappings
- ✅ Improved error handling
v1.0.0
- Initial release
- Basic MCP tools
- Streamable HTTP transport
Live MCP Server: https://app.cogniz.online/mcp
Get Your API Key: https://cogniz.online/dashboard
Need Help? Open an issue or contact support@cogniz.online
Made with ❤️ for the AI community
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。
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。