Skip to main content

How Glyphs MCP works

Glyphs MCP lets an AI client such as Codex or Claude Code call tools that run against the Glyphs app you already have open.

It’s designed to feel like: “My assistant can inspect and operate Glyphs safely, with explicit tools.”

The mental model (what runs where)​

  • Glyphs (the app): the UI where your fonts live (glyphs, masters, kerning, spacing, selection).
  • Glyphs MCP plug-in: runs inside Glyphs and bridges GlyphsApp APIs to MCP tools.
  • Local MCP server: an HTTP endpoint on your machine (localhost) that exposes those tools.
  • Your AI client: connects to the endpoint and calls tools like list_open_fonts, review_spacing, generate_kerning_tab, etc.

Endpoint basics​

By default the endpoint is:

http://127.0.0.1:9680/mcp/

Two common “gotchas”:

  • If you open that URL in a browser, you’ll typically see a small discovery JSON response.
  • MCP clients connect using Streamable HTTP (SSE), so “browser works” does not always mean “client is configured correctly”.

If you’re stuck, see Troubleshooting.

What “tools-first” means​

You’ll usually get the best results by asking the assistant to use dedicated tools:

  • Good: “Call review_spacing and summarize outliers; then do a dry run with apply_spacing.”
  • Good: “Open a kerning proof tab with generate_kerning_tab.”

Use execute_code / execute_code_with_context when you really need a multi-step workflow that doesn’t have a dedicated tool. For mutating scripts, prefer execute_code_with_context so the work stays anchored to an explicit glyph/layer context. Glyphs undo is glyph-scoped: large glyph edits should still use glyph.beginUndo()/endUndo() or layer.beginChanges()/endChanges(), and master/global changes are not guaranteed undoable.