ARTICLE

n8n 2.35+: Agents preview, MCP, and one-line Docker — where Bubble should stop

Vitalii Doroshenko Published

When launching an MVP with no-code and vibe-coding, it is tempting to overload one platform. Here is an honest breakdown of n8n 2.35+ updates and where Bubble should remain your core product surface.

When non-technical founders set out to build an MVP, speed to market is everything. You need to validate assumptions, gather early user feedback, or present a working prototype to investors without drowning in backend infrastructure. With modern no-code platforms and vibe-coding practices, there is a recurring temptation to push every capability into a single environment. The release of n8n 2.35+ offers a textbook opportunity to establish a practical architectural boundary.

What Bubble Does Best

Bubble is a proven full-stack visual application builder. In a lean testing cycle, its role is well-defined:

  • User Interface & Client Experience: rapid assembly of interactive web layouts, customer dashboards, and sign-up funnels.
  • Data Layer & Privacy Rules: a managed relational database where record-level visibility is handled directly within visual settings.
  • In-App Automation: user-facing interactions such as checkout flows, transactional emails, database triggers, and structured external requests via the API Connector.
  • Zero-DevOps Hosting: instant web deployment without managing containers, servers, or reverse proxies. A closer look: The Node.js/PHP CVE and Bubble’s.

If you need a reliable product where early adopters sign in, interact with data, and pay for value, Bubble is the ideal tool for the job.

What n8n 2.35+ Introduces

The community announcement around the n8n Assistant update and 2.35+ focuses on three core areas:

  1. Agents (Preview): prompt-assisted creation of autonomous agents inside the Assistant, suggesting system instructions, tools, and skills. Unlike traditional node-based workflows, Agents represent a distinct architecture supporting execution channels, schedules, sub-agents, and knowledge bases. Availability is rolling out gradually; self-hosted installations require version 2.32.3+ (Beta) and the environment flag N8N_ENABLED_MODULES=agents.
  2. MCP (Model Context Protocol): support for connecting servers from the MCP registry into the Assistant for real-time development context, alongside an instance-level MCP server (from version 2.34.0) that exposes workflows and agents to external AI development tools like Cursor and Claude Desktop.
  3. One-Line Self-Hosting: a simplified install script (curl -fsSL https://get.n8n.io | sh) for Docker and docker compose v2. This setup bundles SQLite, an internal execution sandbox, and SearXNG search, letting users connect their own LLM API keys (Anthropic, OpenAI, OpenRouter, or OpenAI-compatible endpoints) directly in the web UI.

The Operational Reality: Respecting Preview Status

To keep an early-stage startup resilient, it is crucial to avoid building business-critical dependencies on experimental software. Official n8n documentation explicitly marks both Assistant and Agents as Preview, cautioning developers not to rely on agent executions or agent-exposed MCP tooling in mission-critical production workflows.

Key constraints to consider before integrating:

  • One-line Docker is an evaluation setup: the bundled script targets fresh machines only. Its packaged code sandbox is built for experimentation (official docs point to Daytona for isolated production setups), and npm/npx deployments are officially being retired for new setups ahead of n8n 3.0.
  • Credits and Costs: distinguish between Assistant AI credits and workflow Gateway credits, while keeping API token usage from external model providers in check.
  • No Native Bubble MCP: while n8n provides built-in instance-level MCP exposure, exposing Bubble endpoints to MCP is currently an engineering integration pattern (relying on API Workflows paired with community tools such as mcp-remote or custom proxy servers). A closer look: Community AI Toolkit for Bubble:.

Practical Takeaway for Builders

When testing an MVP, keep your stack modular:

  1. Retain Bubble as your product surface: manage UI state, customer profiles, relational records, and primary permissions inside Bubble.
  2. Use n8n as an auxiliary integration engine: offload complex background jobs, multi-step LLM chains, or external integrations via webhooks and the API Connector.
  3. Isolate experimental features: explore n8n Agents and local Docker deployments to experiment with agentic flows, but keep customer-facing MVP guarantees anchored to stable, deterministic endpoints. A closer look: AI Coding Agents vs. Assistants.

Related articles

← All articles