Compare compiled fonts

Glyphs MCP 2.0.2 adds compare_fonts for compiled static and variable TTFs. Diffenator is deferred to a future release and is not offered in Setup in 2.3.0. A maintained fork is planned to replace youseedee with an independently written reader for bundled Unicode data, preserving offline comparisons. Before redistribution, that fork must pass dependency and license checks, comparison tests, signing, notarization and supported-macOS qualification. The comparison protocol is retained, but comparisons require a separately qualified runtime; check availability before starting a job.

Choose the inputs

Select an explicit previous release or retained export as the baseline. Export the candidate with matching settings using the existing export workflow, then publish that export to a new directory. Export authorization and font saving remain separate. Git source reconstruction and standalone proofs are deferred.

Call get_status and require font.compare.diffenator.v1 in comparisonCapabilities. comparisonWorker explains runtime availability. A disconnected native bridge does not prevent comparing existing font binaries.

{
  "baseline_files": ["/absolute/releases/previous/Regular.ttf"],
  "candidate_files": ["/absolute/exports/candidate/Regular.ttf"],
  "options": {"styles": "instances"}
}

Each side accepts 1–32 existing absolute TTF paths, with a 32 MiB limit per file. OTF, webfont containers and Glyphs sources are unsupported inputs. Options are:

Option	Behavior
styles	instances by default; masters or cross_product for one variable TTF on each side.
filterStyles	Optional regular expression selecting style names.
userWordlist	Optional absolute .txt or .csv file, at most 1 MiB. CSV supports script, language and OpenType feature settings.

Named styles must match on both sides; duplicate or unmatched styles are rejected instead of silently omitted. Cross-product comparisons use candidate axis min/default/max locations, require compatible baseline ranges and axis tags, and are limited to 256 locations. Custom words help cover names and optional features that default wordlists may miss. See Diffenator documentation.

Run and review

Retain the returned job ID. get_job reports input snapshotting, comparison and report verification. Inputs are copied privately and hashed; later edits to the original files do not alter the comparison. Cancellation through discard_job stops the worker and its child processes.

A ready job returns an HTML entryPoint, a report and a verified file manifest. The report records input hashes, runtime revision, comparison options, matched locations and measured glyph/word/table differences. Export settings remain unknown unless retained with the original exports. Counts describe the chosen coverage; completion does not approve the differences.

Use your client's local file tools to open the HTML report. To retain it, call accept_job(job_id, destination="/absolute/new-report-directory"). Publication verifies the snapshots and report bytes and creates a new directory. It refuses existing destinations. apply_job cannot apply a comparison result to a font.

Distribution status

The comparison tool is included in the stable release. The optional runtime's catalog entry stays unavailable until signed, notarized assets and architecture acceptance are recorded. Setup explains this state; it does not install an ambient CLI or download packages during app startup. Browser screenshots are deferred.