n8n Assistant: Agents Preview, MCP, and One-Line Docker 2.35+ — What to Wire into Bubble

When launching an MVP on Bubble without deep engineering skills or a dedicated dev team, speed to market is everything. Non-technical founders looking to test hypotheses, get early customer feedback, and secure traction often rely on Bubble for the UI and data layer, pairing it with n8n via webhooks or HTTP requests for flexible automation logic. Same topic: n8n Agents and MCP next.

The September 2026 n8n Assistant update (streamlined in 2.35+) introduces three components that directly accelerate this workflow:

1. Agents (Preview): First-class artifacts combining models, system instructions, and tool skills. The n8n Assistant can now suggest prompts, skills, and tools through natural language chat to scaffold agent behavior quickly.
2. MCP Registry Integration for the Assistant: Connect external servers from the Model Context Protocol (MCP) registry directly inside the Assistant chat interface, allowing the builder to fetch context and orchestrate tasks while constructing workflows.
3. One-Line Docker Setup (2.35+): Running curl -fsSL https://get.n8n.io | sh bootstraps an n8n Docker Compose environment with bundled sandbox execution and SearXNG search preconfigured. The instance-ai module is enabled by default on Docker starting in 2.35. Same topic: n8n 2.35+: Agents, MCP, and.

What Changes for Bubble Builders

It is important to be clear about the integration boundary: the digest updates do not introduce a proprietary new Bubble plugin. The Bubble-to-n8n transport layer remains standard REST: Bubble API Connector / Backend Workflows trigger n8n Webhook or HTTP Request nodes. The real improvement is operational velocity on the n8n side: founders can vibe-code complex multi-step automations and AI agent routines through conversational prompts rather than wiring every node by hand.

Operational Caveats and Guardrails

• Preview status: Agents and Assistant MCP tools are in preview. Workflows must be inspected and validated before serving live customers.
• BYOK requirement: The one-line installer prepares the containers, but the Assistant remains dormant until you add a model provider key (Anthropic, OpenAI, OpenRouter, or OpenAI-compatible) in settings or environment variables.
• Docker-only path: The streamlined self-hosted Assistant setup is supported on Docker; npm installations are not supported for this deployment pattern.
• Sandbox environments: The bundled sandbox in the one-line setup is meant for local testing and quick evaluation. Production-grade isolation guidance points to Daytona.
• Execution constraints: Queue mode is currently not supported for Agents. Worth reading next: n8n Assistant 2.35+: Agents, MCP,.

For no-code founders building rapid MVPs, these updates drastically shorten the setup cycle: one command deploys the workflow engine, and conversational assistant tooling scaffolds the business logic that connects to Bubble.