n8n Assistant 2.35+: preview Agents, MCP, and one-line Docker next to Bubble

When launching an MVP without an engineering background, your primary focus is speed: validating the concept, getting early customer feedback, or presenting traction to investors. Bubble is an exceptional no-code tool for building interfaces, managing authentication, and iterating on your data model. However, running multi-step AI chains, document parsing, or scheduled web scrapers directly inside Bubble workflows can quickly drain Workload Units (WU) or trigger execution limits. A closer look: AI Coding Benchmarks: How to. Worth reading next: Using an AI Coding App. Worth reading next: Free ai coding tools next.

For solo founders, running a self-hosted automation engine like n8n as a companion «sidecar» backend has become a standard architectural pattern: Bubble handles the frontend, while n8n manages heavy logic. With release 2.35+, n8n has introduced several features that make setting up this stack significantly faster for vibe coders and non-technical builders. Same topic: Claude Code 2.1.287: What Non-Technical. Worth reading next: Who Should Build Your Bubble.

What n8n 2.35+ Brings to the Table

Several updates in this release directly impact rapid MVP development:

1. Renamed to n8n Assistant: Replacing the earlier AI Workflow Builder, n8n Assistant is a canvas-native helper. You can build, modify, test, and debug workflows using natural language prompts. It prompts you for required credentials and helps troubleshoot broken nodes directly on the canvas.
2. One-Line Docker Setup: Instead of deciphering complex server guides, self-hosting is now streamlined via curl -fsSL https://get.n8n.io | sh. It requires Docker with the docker compose v2 plugin. The installer creates a local ./n8n directory running on port 5678 with SQLite and bundled sandbox and web-search services.
3. Autonomous Agents (Preview): The assistant can scaffold agentic architectures where an LLM dynamically selects tools and branches based on input data rather than following rigid, static paths.
4. Three Layers of MCP: Model Context Protocol integration appears across the stack: connecting external MCP servers from a registry directly inside the Assistant chat, exposing your n8n instance as an MCP server via settings or N8N_MCP_ACCESS_ENABLED, and native workflow nodes (MCP Server Trigger and MCP Client Tool).

What a Solo MVP Founder Can Test Today

• Vibe Coding Backend Endpoints: Instead of configuring JSON payloads and complex routing rules manually, you can tell n8n Assistant: «Create a webhook that accepts text from Bubble, summarizes it with OpenAI, and returns a JSON response». The assistant drafts the workflow on your canvas.
• Cost-Efficient Execution: Running heavy AI transforms on your own VPS keeps Bubble Workload Units lean, giving you predictable monthly infrastructure costs during early validation.
• Bring Your Own API Keys: On self-hosted Community instances, you are not bound to platform credit systems; you authenticate directly with your model provider (OpenAI, Anthropic, etc.).

Critical Gotchas and Preview Constraints

To build a dependable MVP, keep these technical boundaries in mind:

• Preview Means Always Verify: n8n Assistant and Agents are explicitly in preview. They can misconfigure node parameters or hallucinate settings. Always test generated workflows with sample data before routing real user requests through them.
• The Bubble API Connector Timeout (Error 524): This is the most frequent stumbling block. Bubble's API Connector enforces strict timeout boundaries (~30 to 45 seconds). If your n8n agent runs multiple LLM calls and takes 60 seconds, Bubble drops the connection and shows a failure. Solution: Use n8n’s Respond to Webhook node right away to return an immediate 200 OK or confirmation token, then deliver the final output back to Bubble asynchronously using Bubble’s Workflow API (Backend Workflows).
• Docker-Only Requirement: npm-based installations are not supported for Assistant. n8n documentation notes a broader shift away from npm toward Docker leading into n8n 3.0 (scheduled around October 2026).
• Manual Agents Flag: Agent capabilities are rolling out gradually. Enabling them on self-hosted setups often requires setting N8N_ENABLED_MODULES=instance-ai,agents. Note that self-hosted Enterprise support for agents is not ready yet, queue mode is unsupported, and channel integrations require a publicly accessible WEBHOOK_URL.
• Sandbox Hardening: The bundled sandbox included with the one-line installer is designed for quick prototyping. Production deployments running arbitrary code should follow official guidance toward dedicated execution environments like Daytona.

Practical 5-Step Action Plan

1. Deploy the stack: Run curl -fsSL https://get.n8n.io | sh on a local machine or a cloud VPS with Docker v2 installed.
2. Configure your LLM: Add your model provider API key in the n8n UI credentials or .env file.
3. Prompt the Assistant: Request a workflow with a Webhook node, an AI step, and a Respond to Webhook node.
4. Hook into Bubble: Add a new API call in Bubble's API Connector set to Action / POST, pointing to the n8n test webhook URL.
5. Verify and decouple: Test synchronous responses for quick lookups; if processing takes longer than 15 seconds, switch to an asynchronous callback using Bubble's Backend Workflows.