Your Design System MCP Shouldn't Be Costing You Tokens
Tim Gremore · 10 November 2026
Read the originalTim Gremore of the agency Headway says a design system's AI costs depend on how often its context loads; in an Atlassian test, a smaller file cost more.
Tim Gremore, development lead at Headway, an agency that builds design systems, answered a question from a live session about what AI coding agents cost on a design system. A team had got the output it wanted but spent many credits getting there. Gremore argues that most teams pay for context their agent never asked for.
- A DESIGN.md file is read in full on every turn, so a file covering every component loads all of them whether the agent builds a button or a table.
- In a test Atlassian published, an 80 KB DESIGN.md used 7.21 million tokens to build one log-in screen. An MCP server holding about 2.5 MB of guidance, fetched in slices, used 3.75 million.
- Headway does not point agents at Storybook's MCP, which it found expensive. It ships a custom MCP server as a local npm package, with no network call or login.
- Gremore recommends keeping intent and visual rules in DESIGN.md, having the MCP server read from it, and using fixed code rather than AI for work whose rules are known.
A design system team giving agents its rules could measure what one agent run costs before choosing between a single file and a server that answers narrow questions. The piece is an agency's advice and cites one Atlassian test on one task.
Compare the tokens an AI agent spends building one screen when it reads one long design rules file in full and when it fetches only the rules it needs.
Give AI coding agents a small local tool that answers narrow questions about the design system, such as which properties a component takes, instead of loading every component's documentation on every turn.
Derived by Working Surface from the article; more in the Playbook. Source line: You’re paying to render a human experience so a machine can read it.
13 November 2026: AI agents earn more freedom when a team limits what they can reach, from data to design components, and adds a check that does not depend on a glance.