n8n Agents & MCP with Bubble: Docker 2.35+ Guide for Fast MVPs

When launching an MVP to test an idea, gather early customer feedback, or demo traction to investors, the Bubble + n8n stack has become a staple for indie makers and vibe-coders. Bubble provides the fast responsive interface, user authentication, and relational database, while n8n handles third-party automations and heavy integrations. Same topic: Composio vs official Bubble MCP:. A closer look: AI Coding Benchmarks: How to. A closer look: AI Coding Agents vs. Assistants. Same topic: n8n 2.35+: Agents, MCP, and.

The recent n8n release around version 2.35+ introduces notable changes: the rebranding to n8n Assistant, the rollout of an Agents (preview) module, native MCP (Model Context Protocol) touchpoints, and a one-line Docker installation script. For a non-technical founder building at speed, the key question is simple: what does this actually change, and where should you draw the boundary between Bubble and n8n? A closer look: n8n Assistant 2.35+: preview Agents,.

What n8n 2.35+ Actually Delivers

Public releases and documentation point to three core updates:

One-Line Self-Hosted Setup (Docker)
Fresh self-hosted installs can now leverage curl -fsSL https://get.n8n.io | sh. The script configures Docker Compose, a code execution sandbox, and a local search utility (SearXNG).
Version note: Community release threads mention 2.35+, while official blog posts cite 2.36+ as the baseline for self-hosted Assistant features. Regardless of the exact sub-version, the assistant services start in a ready state, but you still need to provide your own model provider API key (OpenAI, Anthropic, etc.) via the UI or .env before the assistant is functional.
n8n Assistant & Agent Scaffolding (Preview)
Renamed from "AI Assistant", the n8n Assistant focuses on building, modifying, testing, and debugging workflows on the canvas using natural language. It also supports scaffolding agents with instructions, tools, and skills. However, the standalone Agents feature remains a preview module (N8N_ENABLED_MODULES=...,agents), does not support execution queue mode, and requires human oversight.
Two Layers of MCP Support
- Assistant Registry Search: The assistant can search and connect tools from the MCP registry to assist in building workflows (controlled via N8N_INSTANCE_AI_MCP_CONNECTIONS_ENABLED).
- Instance-Level MCP Server: n8n can expose selected workflows and agents directly to external MCP clients (toggleable via N8N_DISABLED_MODULES=mcp).

The Architectural Split: Where to Keep Your Logic

When prototyping rapidly, it is tempting to offload entire workflows and decision-making to an AI agent in n8n. However, maintaining app stability and user trust requires a clean boundary between Bubble and n8n.

Keep strictly in Bubble:

User Identity & Session Management: User signup, login states, role assignments, and permission checks must stay inside Bubble's native user system.
Privacy Rules & Sensitive Data: Bubble enforces row- and field-level security natively. Customer records, payment history, and private user files should never be exposed over unauthenticated endpoints.
Core State Mutations: When an action changes critical business state—such as updating a subscription status or confirming an order—the final database update must be executed by a Bubble Backend Workflow that validates the request.

Delegate to n8n:

AI Reasoning & Agent Execution: Multi-turn LLM chains, agent loops with tool usage, and web-grounded research are well-suited for n8n's preview Agents module.
MCP Tool Orchestration: Connecting external data sources, niche APIs, or code interpreters through the Model Context Protocol is significantly easier to maintain in n8n than building custom API connectors in Bubble.
Heavy Background Jobs: Multi-step batch processing, data enrichment, or synchronization across third-party tools should run outside Bubble to conserve Bubble Workload Units (WU) and maintain a fast front-end experience.

A Pragmatic MVP Integration Pattern

To move fast without compromising security, use this straightforward integration loop:

Trigger from Bubble: The user clicks an action in the UI (e.g., "Generate Report"). A Bubble workflow uses the API Connector to send an HTTP POST to an n8n webhook, passing an opaque record ID, user identifier, and parameters.
Orchestrate in n8n: The n8n workflow executes the agent, gathers context via MCP servers or API nodes, and generates the required output asynchronously.
Authenticated Callback to Bubble: n8n posts the structured result back to a Bubble Backend API Workflow.
Security rule: Protect the Bubble backend endpoint with an API token or bearer secret. Avoid casually enabling "Ignore privacy rules" on public workflows unless strict conditional checks are applied inside the Bubble workflow itself. A closer look: Retool’s NL PostgreSQL/REST Resources: What.

Honest Risks for Founders and Builders

Preview Instability: Both n8n Assistant and Agents are preview features. APIs, node configurations, and behavior may change between minor versions. Do not deploy uninspected agent-generated workflows directly into business-critical production paths.
Self-Hosting Constraints: While curl | sh simplifies deployment, running self-hosted agents with knowledge bases requires Daytona, and agents cannot currently run in multi-instance queue mode. For pure no-code teams, managed n8n Cloud or standard n8n Docker containers may require less DevOps maintenance.
API Token Consumption: On self-hosted setups, you pay the LLM providers directly for every token consumed by the Assistant and runtime Agents. Monitor token usage closely during rapid testing.

Bottom Line: n8n 2.35+ significantly accelerates how quickly you can scaffold agentic tools and connect them via MCP. But treat n8n as the intelligent execution engine, while Bubble remains the single source of truth for your users, data security, and core business rules.