Skills and optional plugins

Skills tell an agent how to use the supported workflows. They do not add tools to the server or grant permission to edit a font.

Opening an existing native font uses open_document and the shipped glyphs/references/font-opening.md guidance. It requires a coordinated bridge and sidecar with document.open.v1; it does not require a saved baseline or script.

The package at plugins/glyphs-mcp contains manifests for Codex, Claude Code, Cursor, and GitHub Copilot CLI. All four manifests use version 2.4.0 and share the same skills. Current v2 Setup configures Codex and Claude Code with MCP plus managed skills, configures Claude Desktop's MCP entry, and installs the complete local Cursor bundle at ~/.cursor/plugins/local/glyphs-mcp. GitHub Copilot CLI remains a package-supported host without a v2 Setup card.

The agent plugin is optional for using the twenty-one data/action tools and optional card-display tool; a manual MCP connection can use the same server catalog. Install All nevertheless includes all four supported Setup connectors by default, even if a host is not detected. For 2.4.0, use the matching local package or its published equivalent. Published 2.0.0 packages remain matched to the released 2.0.0 runtime. Configuration patching creates backups, preserves unrelated entries and custom options, and changes only Glyphs MCP. Cursor update and removal require an installer-owned bundle whose version and hash still match its receipt; modified or unowned content becomes a visible conflict. Public marketplace commands resolve published state; verify that the installed package version matches your runtime before using its workflow instructions.

Included skills

The current skills/manifest.json contains twelve managed skills.

Skill	Use
glyphs	Verify the specific connection when needed; reuse its context and document ID.
glyphs-mcp-spacing	Prepare and review sampled spacing suggestions.
glyphs-mcp-kerning	Review explicit collision-kerning pairs.
glyphs-mcp-italic-first-pass	Prepare a guarded first-pass slant.
glyphs-mcp-master-compatibility	Review start-node correspondence; full master compatibility repair is not advertised.
glyphs-mcp-outlines-docs	Inspect bounded outline information and native documentation.
glyphs-mcp-development	Develop reusable scripts or plugins in the workspace.
glyphs-mcp-scripting	Prepare reviewed scripts for native execution.
glyphs-mcp-opentype-features	Review and debug native Glyphs 4 feature source, compilation and exported behavior.
glyphs-mcp-hinting	Inventory hints and compare PostScript, CFF2 or TrueType rendering before scoped repairs.
glyphs-mcp-release	Prepare and validate repository releases.
glyphs-mcp-maintainer-feedback	Turn a reproducible issue into useful maintainer evidence.

Use the general glyphs skill when unsure which workflow fits. Workspace development can start without an open font. Live jobs require a ready server/bridge and an explicit scope. Script jobs can run on unsaved or dirty fonts with pre-run memory recovery; other operations retain their advertised baseline requirements.

The v1 guide retains its marketplace and skill instructions. Do not reuse v1 tool calls with the v2 server. See the migration guide.

OpenType source work uses glyphs-mcp-opentype-features for native Glyphs 4 scripts, compilation diagnostics and exported behavior checks. This skill adds no additional public tool: native source edits use the existing python_script workflow when script.native.v1 is available.

Preserved conflicts and retired private skills

Unchanged installer-owned skills refresh automatically. User-edited or unowned folders remain preserved-conflict with their exact path and a repair action. Review them, then use Replace preserved skills (backup) when replacement is intended. Exact backups and restoration coordinates go in the sibling glyphs-mcp-skill-backups folder outside skill discovery, not beside selectable SKILL.md files. The installer log gives each complete backup path.

Recognized retired private icon-font, LitSquare, color-font, variable-font, production-audit, Unicode-semantics and export-validation folders are archived by that action; they are not replaced by newly supported audit tools. Separate v1 instructions are preserved. The current entry explains bounded reads, separately scoped native coding, and unsupported specialized tasks.

A skill requiring get_server_info and apiMajor == 2 can be an obsolete private instruction, not proof that a current lean runtime needs updating. Check its actual path and connection catalog. The installer does not edit plugin caches or source worktrees. Refresh the client's skill discovery after changes; an existing task may retain old instructions. Do not assume hot reload.

Skills catalog

Open Skills below Setup to browse included, curated and reviewed community workflows. Companion Plugins and AI Agents open focused pages that share the same operations and status with the complete Setup overview.

Included means shipped with this app; it is separate from installation status. Compact text cards show each workflow’s goal and whether it is included, installed or available. Open Details for client copies, instructions, source and licensing. Included skills link to AI Agent Setup; additional skills offer installation, explicit updates and removal from their detail view. There is no global on/off switch: skill discovery belongs to each client and does not enable or disable the corresponding MCP tools.

Supplemental Codex/Cursor skills share ~/.agents/skills; existing copies in ~/.codex/skills or ~/.cursor/skills are detected and updated in their actual location. Claude Code uses ~/.claude/skills. Choose Project scope and select a folder to install into .agents/skills or .claude/skills inside that project. Installation does not commit project files or alter existing bundled-skill locations.

Import from GitHub accepts a public repository URL with separate directory and optional revision fields. It resolves an immutable commit, lets you choose from multiple skills and keeps the import personal and unreviewed. Inspect its instructions, file list, license and dependencies before installation. Private repository authentication and Gist import are not included in this release.

Updates apply only when selected. Details offers Install/Update and Remove for each supported agent. Edited or unmanaged copies require explicit confirmation and a backup before replacement or removal. Linked folders are labelled: Update replaces only the link with a managed copy after confirmation; Remove detaches the link. Both preserve its external source folder. Competing copies in other Codex/Cursor locations are preserved. Backups stay outside skill discovery with exact restoration paths. Supplemental receipts are independent of bundled Setup receipts.

Export ZIP produces a complete skill folder for guided upload to ChatGPT or Claude where account permissions allow. Export is not installation. Upload and enable the package in the host and configure its dependencies separately; uploading instructions alone does not make Glyphs on this Mac accessible to hosted execution. The app does not execute downloaded scripts or install MCP connections/dependencies as part of skill installation.

The contribution repository accepts skill folders and pinned external GitHub entries through reviewed pull requests. Its template, submission forms and validator support a first contribution. Curated entries are selected by maintainers; Community entries are reviewed submissions. The app includes an offline snapshot and retains its last valid catalog if a refresh fails. Optional skills remain outside Install All.

Optional development skills in Setup

Select Include Glyphs MCP development skills in Connections to add script/plugin development and release skills to Codex and Claude Code during setup. The checkbox starts unchecked and remembers your choice. Existing development skills stay in place, and local development or release copies do not make connection setup fail. The development skill is also available in the Skills catalog.