Gemini 4 Argon + n8n: Assistant, MCP, Dashboard

By n8n

Categories: Tools, Product

Summary

Gemini 4 Argon launched with aggressive pricing ($2 per million input tokens, 95% cached input discount) and lowest prompt injection attack rates in benchmarks. n8n now integrates it across assistants, MCP, and workflows—enabling AI agents to build automation, draft responses, and create dashboards end-to-end.

Key Takeaways

  1. Gemini 4 Argon achieves 91.9% on Vibe Code Bench and lowest attack success rate on Gray Swan's prompt injection benchmark—critical for AI agents reading untrusted input in production workflows.
  2. n8n assistant auto-generated a complete quote-request workflow (lead scoring, reply drafting, data table storage, Slack alerts) in single prompt, reducing build time from hours to minutes.
  3. MCP integration via anti-gravity ID enables Gemini agents to access n8n tools through CLI—install n8n skills with `npx skills add n8n.io/skills` to provide API context and reduce token waste.
  4. Use instance-level MCP token generation in n8n settings to connect external agents; paste JSON config directly into anti-gravity MCP settings for seamless tool discovery and execution.
  5. Pricing model ($10 per million output tokens vs industry standard) combined with 95% cache discount makes long-horizon workflows economical—especially for agents making multiple LLM calls.

Related topics

Transcript Excerpt

Gemini for Argon has finally dropped and it's available today in n8n. In this video, we'll see how you can connect it with n8n, first through our assistant, then using MCP through anti-gravity, and also, of course, actually having it as a part of the workflow. But first, let's take a look at Google's own benchmarks. Three of them matter for building an n8n. Vibe Code Bench at 91.9%, Deep Software Engineering at 77.9%, and Automation Bench at 51.3%. across complex long horizon workflows, which is what we're about to test. At launch, the introductory API pricing is $2 per million input tokens and $10 per million output with 95% off on cached inputs. Let's first start building with the n8n assistant. By the way, I'm running this locally and to make sure that your assistant is using the right …

More from n8n