cleancontext.blog
notes on AI

Context,
agents
and automation
with LLMs.

I write about what I learn building with LLMs — from context engineering to agents and automation — and publish the tools I use along the way.

Latest post

see all →
n8n · MCP

A 100+ node n8n workflow costs a fortune in tokens for an LLM to read directly

I tested both ways of reading a workflow in production: straight through MCP, and through a deterministic reducer. One hit a token ceiling error. The other cut 84%.

Read

Featured tool

see all →
runs in the browser

n8n Workflow Reducer

Paste an n8n workflow export and see the reduced skeleton — business logic separated from storage metadata, side by side with the token savings.

Open