Integrations (MCP)
Connect external data sources to DAISI using the Model Context Protocol. Your data is automatically synced into Drive and indexed for RAG-powered AI search.
What is MCP?
The Model Context Protocol (MCP) is an open standard for connecting AI systems to external data sources. DAISI's MCP integration lets you register remote MCP servers — databases, APIs, file systems, SaaS platforms — and automatically sync their resources into your DAISI Drive account.
Once synced, files are indexed into the vector database and become searchable via natural language queries. This powers RAG (Retrieval-Augmented Generation), allowing AI conversations to reference your private data while respecting access controls.
Getting Started
- Navigate to Integrations — In the DAISI Manager, click "Integrations" in the top navigation bar.
- Add a Server — Click "Add Server" and fill in:
- Name — A friendly name for the data source
- Server URL — The MCP server's endpoint URL (SSE or HTTP)
- Authentication — None, Bearer Token, or API Key as required by the server
- Target Repository — (Optional) Which Drive repository to store synced files in. Leave blank to auto-create one.
- Sync Interval — How often to sync (5 minutes to 24 hours)
- Review Resources — After registration, the system discovers available resources on the server. Toggle which resources you want to sync.
- Sync — Syncing happens automatically on the configured interval. You can also click "Sync Now" for an immediate sync.
How Sync Works
When a sync runs:
- A host on the DAISI network connects to your MCP server via SSE
- It lists available resources and reads the ones you've enabled
- Each resource is uploaded to your Drive repository
- Content is automatically chunked and indexed into the vector database
- Files that haven't changed (same content hash) are skipped to save bandwidth
Synced files carry a SourceUri linking them back to the original MCP resource,
making it easy to trace where data came from.
Authentication
DAISI supports three authentication modes for MCP servers:
| Type | Description | Use Case |
|---|---|---|
| None | No authentication | Public or localhost servers |
| Bearer Token | Authorization: Bearer <token> header | OAuth-protected APIs |
| API Key | X-API-Key header | API-key-protected services |
SDK Usage
You can also manage MCP servers programmatically via the .NET SDK:
var mcpClient = new McpClientFactory(clientKeyProvider).Create();
// Register a server
var response = await mcpClient.RegisterServerAsync(
name: "My Database",
serverUrl: "https://mcp.example.com",
authType: McpAuthType.McpAuthBearer,
authSecret: "my-token",
targetRepositoryId: null,
syncIntervalMinutes: 60
);
// List servers
var servers = await mcpClient.ListServersAsync();
// Trigger sync
await mcpClient.TriggerSyncAsync(serverId);
See the McpClient SDK Reference for the full API.
MCP Data Access
In addition to connecting to MCP servers, the DAISI Manager also exposes an MCP server endpoint so you can access your account data (credits, earnings, spending, transactions) from any MCP client like Claude Desktop or VS Code Copilot.
Access Control
Data synced from MCP servers is stored in Drive repositories with the same access controls as any other Drive files. When you search via vector search, results are automatically filtered to repositories you have access to — private data stays private even through semantic search.
See the DriveClient SDK Reference for vector search with repository filtering.