Unified Connections

Connections are PuppyOne's unified abstraction for managing every external integration.

What is a Connection

In PuppyOne, whether you want to sync data from Notion, create an AI agent, expose an MCP endpoint to Cursor, or mount a local folder, they are all represented as connections.

All connections are stored in the same connections table and distinguished by fields like provider and direction. That gives you one consistent interface for every external integration instead of separate models and workflows.

Connection types

Type	Description	Direction	Typical Use Case
Sync Source	External data source sync	Pull	Pull data from Notion, GitHub, Gmail, and more into Content Nodes
Agent	AI agent	—	Bind a model, system prompt, and tools for chats and automation
MCP Endpoint	MCP protocol endpoint	Bidirectional	Expose data access to clients such as Cursor and Claude Desktop
Sandbox	Code sandbox	—	Run code in isolated Docker or E2B containers
Filesystem	Local folder	Bidirectional	Real-time sync between a local folder and the cloud through OpenClaw

Sync source

Automatically sync external sources such as Notion pages, GitHub repositories, and Gmail messages into your Project as structured Content Nodes.

Supported sources include Notion, GitHub, Gmail, Google Drive, Google Docs, Google Sheets, Google Calendar, Linear, Airtable, file upload, web scraping, and 15+ platforms in total.

Agent

Create an AI agent with a model such as GPT-4 or Claude, a system prompt, and bound tools. Agents can stream over SSE and can also run on schedules.

MCP endpoint

Create an MCP endpoint with its own URL and API key. Any MCP-compatible client, including Cursor, Claude Desktop, and Claude Code, can access your data through it.

Sandbox

Create an isolated execution environment where agents can run code safely. PuppyOne supports both Docker-based sandboxes and E2B cloud sandboxes.

Filesystem

Mount a local folder to PuppyOne through the CLI for real-time, bidirectional sync. Local agents such as Claude Code can work directly against local files while staying connected to your shared context.

Connection properties

Each connection includes these core properties:

Property	Description
`id`	Unique connection ID
`name`	Connection name
`provider`	Provider type, such as `notion`, `github`, `mcp`, or `agent`
`direction`	Sync direction: `pull`, `push`, or `bidirectional`
`status`	Connection status
`access_key`	Access key used in cases like MCP endpoints and agents
`config`	Connection configuration in JSON form, varying by type

Connection lifecycle

Every connection has one of three states:

Status	Description
`active`	Running normally
`paused`	Temporarily paused and can be resumed at any time
`error`	Something is wrong and the connection needs attention, reconfiguration, or reauthorization

Manage connections through the CLI

PuppyOne CLI provides a single conn command for every connection type:

# Add connections
puppyone conn add notion <url>         # Add a Notion data source
puppyone conn add github <repo>        # Add a GitHub repository
puppyone conn add mcp "My Endpoint"    # Create an MCP endpoint
puppyone conn add agent "My Agent"     # Create an agent
puppyone conn add sandbox "My Box"     # Create a sandbox
puppyone conn add folder ~/my-docs     # Mount a local folder
 
# View connections
puppyone conn ls                       # List all connections
puppyone conn info <id>                # View connection details
 
# Manage connections
puppyone conn pause <id>               # Pause a connection
puppyone conn resume <id>              # Resume a connection
puppyone conn rm <id>                  # Delete a connection

Example: Create an MCP endpoint and inspect it

$ puppyone conn add mcp "Product Knowledge Base MCP"
✓ MCP endpoint created
  ID:         conn_abc123
  Name:       Product Knowledge Base MCP
  API Key:    sk_live_xxxxxxxxxxxx
  Server URL: https://api.puppyone.ai/api/v1/mcp/server/sk_live_xxxxxxxxxxxx
 
$ puppyone conn info conn_abc123
  Name:       Product Knowledge Base MCP
  Provider:   mcp
  Status:     active
  API Key:    sk_live_xxxxxxxxxxxx
  Created:    2026-03-09T10:00:00Z

Manage connections through the Dashboard

Open a Project, then click Connections in the sidebar to view and manage everything in one place. Click New, choose a connection type, and follow the guided setup flow.

Best practices

Use one connection per agent - Separate access keys and permission settings make auditing and management much easier
Create MCP endpoints for specific use cases - Different clients or workflows should have their own endpoints and independent permissions
Pause instead of deleting when possible - Temporary connections can be resumed later without recreating everything
Check error states regularly - These often come from expired OAuth credentials or changed configuration

Content Node MCP Protocol