Skip to main content

How Glyphs MCP works

Glyphs MCP lets an AI client call structured tools against the Glyphs app running on your Mac.

The important idea is simple: Glyphs remains the place where the font lives. The agent does not operate a remote copy or scrape the UI. It asks the local MCP server to run explicit tools inside Glyphs.

What runs where​

  • Glyphs app: holds the open fonts, masters, glyphs, layers, kerning, spacing, and current selection.
  • Glyphs MCP plug-in: runs inside Glyphs 3 and bridges GlyphsApp APIs to MCP tools.
  • Local MCP server: exposes a Streamable HTTP endpoint on your machine.
  • AI client: Codex, Claude, or another MCP client connected to the local endpoint.

Default endpoint:

http://127.0.0.1:9680/mcp/

Tools-first behavior​

Ask the agent to use dedicated tools first:

  • list_open_fonts to find the active font context.
  • get_font_masters to identify masters.
  • review_spacing before any spacing apply step.
  • generate_kerning_tab or review_kerning_bumper before kerning changes.
  • docs_search and docs_get for focused Glyphs API lookup.

Use execute_code or execute_code_with_context only when a workflow does not have a dedicated tool or when one focused script is safer than many dependent calls.

State and verification​

Font editing is stateful. A safe agent workflow should:

  1. Read current context.
  2. Confirm exact targets: font, master, glyph, layer, path, node, or selection.
  3. Use the smallest valid tool call.
  4. Dry-run or request confirmation before mutation.
  5. Read back affected state.
  6. Report what changed, what was skipped, and any remaining risk.

Browser vs MCP client​

Opening http://127.0.0.1:9680/mcp/ in a browser can show a small discovery response. That only proves the local server is reachable. MCP clients still need correct Streamable HTTP configuration.

If the browser works but the client does not, see Troubleshooting.