​
MCP Integration

Connect Claude Desktop, Cursor, and VS Code–based clients to Nebula via MCP to store and search memories directly from the assistant.
Prefer web-hosted servers for the simplest setup—no local installation required. Use local servers only if needed for specific environments.

​
Quick start

​
Verify with Inspector

Use MCP Inspector to confirm auth, tools, and health before troubleshooting. 
# Node server under Inspector
npx @modelcontextprotocol/inspector npx @nebula-ai/mcp-server --stdio --api-key your_api_key
If the browser UI loads and the tools are visible/invokable, the server is correctly configured.

​
Environment (Local Servers Only)

  • NEBULA_API_KEY (required): API key for Nebula.
  • NEBULA_API_URL (optional): Custom API endpoint, defaults to https://api.nebulacloud.app.

​
Web-Hosted MCP Servers

Preferred method: Use cluster-scoped URLs for direct connection:
  • URL Pattern: https://mcp.nebulacloud.app/{cluster_id}
  • Authentication: Bearer token via Authorization header
  • Benefits: No local processes, automatic updates, cluster isolation

​
Tools provided

  • add_memory: Store documents or conversation messages
    • content (required): Text content to store
    • role (optional): For conversation messages (user, assistant)
    • parent_id (optional): For conversation threading
    • metadata (optional): Additional metadata dictionary
  • search_memories: Search and retrieve memories
    • query (required): Search query string
    • limit (optional): Maximum results (default: 10)
Note: Web-hosted servers automatically scope to the cluster in the URL—no cluster_id parameter needed!

​
Usage pattern

  • Web-hosted: Each URL is cluster-specific, tools automatically work within that cluster
  • Local servers: Use cluster_id parameter to specify which cluster to operate on

​
Next Steps

More information: