MCP Connection. Expert Skills. Production-Ready n8n Workflows.
Google Antigravity launched in November 2025 and changed what an IDE can do. Autonomous agents that plan, execute, and verify across editor, terminal, and browser. Parallel task execution. Skills that load specialised knowledge on demand.
ag8n makes all of that work for n8n.
The MCP Architecture
Model Context Protocol is the bridge between Antigravity and external tools. ag8n provides a purpose-built MCP server that gives Antigravity structured access to your n8n environment — nodes, properties, operations, documentation, and templates.
The server is forked from czlonkowski/n8n-mcp, one of the most widely adopted n8n MCP implementations. Our fork preserves the core access layer and removes everything that should not be there: all telemetry subsystems, Docker dependencies, and unnecessary overhead. The result is a lightweight server compiled to clean JavaScript.
No cloud dependencies. No data leaving your network. The entire codebase is auditable.
What the MCP connection provides:
- Structured access to 1,000+ n8n nodes (core + community)
- Node properties with 99% schema coverage
- Operation coverage across available actions
- Documentation from official n8n sources
- 2,700+ workflow templates with full metadata
- Community node search with source filtering
The connection alone is useful. The skill pack is what makes it powerful.
Why Raw MCP Access Is Not Enough
Giving Antigravity access to n8n node data is like giving someone a dictionary and asking them to write a novel. They have every word available but no understanding of structure, style, or what actually works.
n8n has nuances that trip up generalist AI models consistently. Display Options change available fields based on selections the agent has not made yet. Expression syntax looks like JavaScript but deviates in ways that cause silent failures. Data structures vary between node types. Configuration patterns that work in isolation break when combined in multi-step workflows.
Independent stress tests on n8n's own AI builder confirm the pattern: fast scaffolding, unreliable configuration. The structure looks right. The details are wrong. And the details are what determines whether a workflow runs or fails at 2am.
This is the gap ag8n's skill pack closes.
How the Skill Pack Works
Antigravity uses a progressive disclosure system for skills. Each skill sits dormant until the agent determines it is relevant to your current request. When you ask for a webhook-triggered workflow, Antigravity loads Workflow Architect and Node Expert. When you need to debug an expression, it loads Expression Mastery. Not all seven at once — only what is needed.
This matters for two reasons. First, it keeps the agent's context window clean, which means faster and more accurate responses. Second, it prevents the "tool bloat" that degrades AI performance when too much information is loaded simultaneously.
Expression Mastery
Covers the complete syntax for n8n expressions, how data flows between nodes, and the specific debugging patterns for the most common failures. The agent writes expressions that work first time because it understands how n8n expressions differ from standard JavaScript.
Workflow Architect
Encodes design patterns for the automation types you build most often: webhook-triggered processing, scheduled data synchronisation, API polling with conditional routing, event-driven multi-branch logic. The agent does not just connect nodes — it follows proven architectural patterns.
Node Expert
The skill that saves the most time. It covers configuration protocols, Display Options behaviour, credential handling, and the output data shapes that change based on operation type. This is the knowledge that generic AI consistently gets wrong and that costs hours in manual debugging.
Validation Debugger
Enforces a structured validate-then-fix loop. The agent checks its own work against known patterns, identifies misconfigurations, and corrects them before the workflow reaches production. No guessing. No hoping it works.
API Operations
Covers interaction with n8n's own API for workflow management, execution triggers, and credential handling — the foundation for automations that manage other automations.
Code Runner
Provides the execution guide for JavaScript and Python within n8n's runtime, covering environment constraints, available libraries, and data access patterns.
Template Strategist
Enables intelligent template discovery and modification from n8n's library of 2,700+ community workflows, so the agent starts from proven foundations rather than blank canvases.
Privacy Architecture
ag8n takes a specific stance on privacy: your n8n instance is your business.
The MCP server runs on your infrastructure. No usage data is collected. No analytics are transmitted. No telemetry exists in the codebase — it was removed at the fork level, not disabled via configuration. The code is compiled from source you can inspect.
For builders running self-hosted n8n specifically to keep operational data private, ag8n maintains that commitment through the entire integration layer.
Getting Started
ag8n requires Google Antigravity (free during public preview) and Node.js. Install the MCP server, add the configuration to Antigravity, and the connection to your n8n instance is live. The skill pack loads automatically through Antigravity's native skill system.
Five minutes from download to your first AI-architected workflow.