Glossa is a translation management system built around AI. You push translation files in whatever format your apps use, Claude translates them following your glossary and rules, and every string is deduplicated across files and apps. Everything you can do in the dashboard is also an API call, an MCP tool and a CLI command, so coding agents can run your localization.

Which file formats are supported?

JSON (including i18next), YAML, Xcode String Catalogs (.xcstrings), Apple .strings and .stringsdict, Android strings.xml, gettext PO, XLIFF 1.2 and 2.0, Flutter ARB, Java .properties, .NET RESX and CSV/TSV. When you push a format it has not seen, the AI writes a reusable template for it from a sample, which you review once.

Which model translates my strings?

Claude Opus. Every prompt includes the string's context, your glossary (brand names are never translated) and your standing rules for that language, and every translation is checked to keep the source's placeholders.

What does deduplication mean in practice?

Identical source strings across every file and app share one unit with one translation per language. Fix "Sign in" in Spanish once and every key that uses it is fixed. When the same text means different things ("Open" the verb and "Open" the store status), the key is split into its own unit, and the review queue asks about merges that look ambiguous.

Can I use it from Claude Code, Codex or Cursor?

Yes. Add the MCP server at /api/mcp with an API key, or install the plugin, and the agent gets one tool per operation: push and pull files, search strings, work the review queue, run audits, add rules. The same operations are available over REST (POST /api/v1/{op}) and the CLI (npx @meterapp/glossa).

How is it different from Crowdin?

Crowdin is built around human translators working in an editor, with machine translation and AI as add-ons. Glossa assumes AI translates and agents review, so the API covers every action, strings dedupe across projects, and audits feed fixes back into rules. Crowdin has a much larger integration catalog, translator marketplace and vendor workflow features that Glossa does not try to match.

How much does it cost?

Glossa is free for one app (2,000 AI translations a month). Pro is $49 a month for 25,000 AI translations, Team $199 a month for 150,000; yearly is two months free. Every plan has every feature, and imports, reviews and edits are free. Write to jonathan@welcome.meterapp.co for Enterprise.

Do I have to move my files?

No. Files stay in your repository. Push the source file from CI or a pre-commit hook, and pull translated files back in the same format and path layout. Comments, key order and file-specific details are preserved where the format allows.

Free for one app

Localization your AI agents can run

Push any translation file. Claude translates it with your glossary and rules, identical strings are shared across every app, and your agents review, audit and fix it over MCP, CLI and API.

Get started Quickstart

A replacement for Crowdin by Meter. See how they compare.

claude · glossa mcp

Work through the Glossa review queue. Split merges whose context differs.

claim_tasks({ "kind": "merge_review", "limit": 10 })

10 tasks · "Open" used by 3 keys: menu.file.open, door.status.open, store.hours.open

menu.file.open is a verb; the other two are adjectives. Splitting it out.

resolve_task({ "task": "8c1e…", "action": "split", "key_ids": ["menu.file.open"], "note": "verb vs adjective" })

Split into a new unit; queued translation for 9 locales.

run_audit({ "locale": "es", "question": "Are we consistent about tú vs usted?" })

Audit queued. Findings arrive as audit_finding and rule_suggestion tasks.

The problem

Translation management was designed before agents could do the work

Crowdin and tools like it are good at coordinating human translators. Most teams shipping apps today want something else: machine translation they can trust, and a workflow their agents can run end to end.

Built for a translator in an editor

Traditional TMS tools center on a human working string by string. Machine translation is a pre-fill step, and the rest of the workflow assumes someone clicks through it.

The same string, translated five times

"Sign in" in your iOS app, Android app and web app are three strings in three projects. Fix one and the other two stay wrong.

Half the product is not in the API

Agents and scripts can upload and download files, but reviewing, approving and fixing quality problems happens in a web UI.

Quality does not compound

A reviewer fixes a mistranslation and the same mistake comes back in next week's strings, because the fix never reached the next prompt.

How it works

From source file to every language in four ops

1push_file
Push any file
Send the source file from your repo, CI or agent. The format is detected; strings are matched against every other app in the workspace.
2translate
Claude translates
New strings are translated into every target locale with their context, your glossary and your rules. Placeholders are checked on every string.
3claim_tasks
Agents review
Low-confidence translations, ambiguous merges and audit findings land in a queue an agent can work through in minutes.
4pull_file
Pull every locale
Get each language back in the file's own format and path layout, with comments kept where the format has them.

Push from your repo

npx @meterapp/glossa push locales/en.json --project web

Pull every locale

npx @meterapp/glossa pull --project web --locale es,fr,de

Formats

Every format your apps already use

12 formats are built in. Push one it has never seen and the AI writes a template for it from your file, which you approve once and reuse forever.

Xcode String Catalog

.xcstringsXcode String Catalogs: every locale, plural variation and comment in one JSON file.

Flutter ARB

.arbFlutter Application Resource Bundles with ICU messages and @key metadata.

JSON (key/value, i18next)

.jsonFlat or nested key/value JSON, including i18next with its _one/_other plural suffixes.

YAML (Rails i18n)

.yml .yamlRails-style YAML locale files with a root locale key and nested scopes.

Apple .strings

.stringsApple .strings files: "key" = "value"; pairs per lproj folder.

Apple .stringsdict

.stringsdictApple .stringsdict plist files for plural rules.

Android strings.xml

.xmlAndroid res/values/strings.xml with <string>, <plurals> and <string-array>.

Gettext PO/POT

.po .potGNU gettext .po and .pot catalogs with msgctxt, plurals and translator comments.

XLIFF 1.2 / 2.0

.xlf .xliffXLIFF 1.2 and 2.0 bilingual files with source, target and notes.

Java .properties

.propertiesJava .properties resource bundles, including Spring messages files.

.NET RESX

.resx.NET RESX resource files with <data> entries and comments.

CSV / TSV (multi-locale)

.csv .tsvCSV and TSV spreadsheets with a key column and one column per locale.

Something elselearn_formatQt .ts, a custom JSON shape, a game engine table: the AI learns it into a reusable template.

Review queue

A review queue an agent can rip through

Every decision that needs judgment becomes a small task with everything needed to decide it and a short list of actions. Agents claim a batch, decide, and resolve. Each resolution runs through the same ops a person would use.

merge_reviewKeys in different apps share text. Same meaning, or split?
translation_reviewThe model was unsure, or a placeholder did not survive.
audit_findingAn audit found a problem and proposes a fix.
rule_suggestionAn audit proposes a rule for every future translation.
glossary_suggestionA term that should be fixed or never translated.
learn_formatThe AI wrote a template for a new file format.

The same queue over REST

curl -X POST https://glossa-meterapp.vercel.app/api/v1/claim_tasks \  -H "Authorization: Bearer gls_…" \  -d '{"kind":"translation_review","limit":20}'
curl -X POST https://glossa-meterapp.vercel.app/api/v1/resolve_task \  -H "Authorization: Bearer gls_…" \  -d '{"task":"…","action":"edit","text":"Iniciar sesión","note":"Use the infinitive for buttons"}'

Deduplication

Fix once, fixed everywhere

Every key points at a unit: one source string with one translation per language, shared by every key in every file and app that uses the same text. When context says the same words mean different things, the key gets its own unit.

ios/en.lproj/Localizable.strings"auth.signin" = "Sign in";android/res/values/strings.xml<string name="sign_in">Sign in</string>web/locales/en.json"header.signIn": "Sign in"

1 unit · "Sign in"3 keys

esIniciar sesióndeAnmeldenjaサインイン

One translation per language, reviewed once. split_key separates a key whose context differs; merge_units joins near-duplicates.

Audits and rules

Translations that get better over time

Ask a question about a language and get an answer, a score and a list of fixes. What keeps going wrong becomes a rule or a glossary term, so the next batch does not repeat it.

01 · run_auditAsk"How is our Spanish? Are we consistent about tú and usted?"
02 · audit_findingFindThe audit scores the locale and files each problem with a proposed fix.
03 · resolve_taskFixAn agent or a person applies, edits or dismisses each finding.
04 · add_ruleLearnRecurring problems become glossary terms and rules that every future translation follows.

Every action is an API

48 ops. Dashboard, REST, MCP and CLI all call the same ones.

Each op has one name, one input schema and one handler. The dashboard has no private endpoints: whatever you can click, an agent can call.

Workspace4

Settings, stats and the activity log.

get_workspace update_workspace get_stats list_activity

Projects3

Apps or repos, each with its source and target locales.

list_projects create_project update_project

Files3

Push translation files in, pull any locale back out.

list_files push_file pull_file

Formats4

Detect, list and teach file formats.

detect_format list_formats learn_format approve_format

Strings10

Deduplicated units, translations, merges and splits.

search_units list_units get_unit update_unit set_translation review_translations translate merge_units split_key scan_duplicates

Review queue3

Claim and resolve review tasks.

list_tasks claim_tasks resolve_task

Glossary3

Brand names and fixed terms.

list_glossary upsert_glossary_term delete_glossary_term

Rules3

Standing instructions for every translation.

list_rules add_rule update_rule

Audits3

Per-language quality reviews.

run_audit get_audit list_audits

Jobs2

Background AI work and its status.

get_job list_jobs

API keys3

Workspace keys for CLI, MCP and CI.

list_api_keys create_api_key revoke_api_key

Billing7

Plans, prices, checkout links and the subscription.

get_pricing get_billing create_checkout_link subscribe cancel_subscription resume_subscription create_billing_portal_link

RESTPOST /api/v1/{op}

MCP/api/mcp

CLInpx @meterapp/glossa

Docs/llms.txt · /openapi.json

Integrations

Works where your agents already work

Connect Claude Code, Codex or Cursor to the MCP server, or run the CLI in GitHub Actions on every push.

claude mcp add --transport http glossa https://glossa-meterapp.vercel.app/api/mcp \  --header "Authorization: Bearer gls_…"

# ~/.codex/config.toml
[mcp_servers.glossa]
url = "https://glossa-meterapp.vercel.app/api/mcp"
bearer_token_env_var = "GLOSSA_API_KEY"

{  "mcpServers": {    "glossa": {      "url": "https://glossa-meterapp.vercel.app/api/mcp",      "headers": {        "Authorization": "Bearer gls_…"      }    }  }}

# .github/workflows/translations.ymlname: Translationson:  push:    branches: [main]    paths: ["locales/en.json"]jobs:  sync:    runs-on: ubuntu-latest    steps:      - uses: actions/checkout@v4      - run: npx @meterapp/glossa push locales/en.json --project web        env:          GLOSSA_API_KEY: ${{ secrets.GLOSSA_API_KEY }}

npx @meterapp/glossa loginnpx @meterapp/glossa push locales/en.json --project webnpx @meterapp/glossa pull --project web --locale es,fr,denpx @meterapp/glossa run_audit --locale es --question "Are brand names consistent?"

Full setup, including OAuth sign-in and the plugin, in the MCP guide.

FAQ

Questions

Push your first file

Glossa is free for one app. Sign in, push a translation file, and pull it back in every language.

Get started Read the quickstart