Structured game strings
Manage keys, unlimited languages, context, screenshots, character limits, translation history, branches and more than fifty file formats instead of passing loose text between spreadsheets.
O01O03Phrase
Continuous game-localization infrastructure for strings, language assets, QA, delivery and external agents.
Evidence statusPublic-evidence reviewMedium confidenceLast checked Sep 15, 2026By MakeGameWithAI
PRODUCT OVERVIEW
What Phrase is—and how it works.
Phrase is a localization platform, not one translation model. For a game team, Phrase Strings is the source-content layer: it stores keys, locales, translations, metadata, screenshots, character limits, branches and history, while translation memory, term bases, jobs, suppliers and QA coordinate the language work around them.
Its AI direction is to place Next GenMT and an AI Translation Agent inside those language assets and controls. That can make drafts more consistent and traceable than sending isolated lines to a general chatbot, but current public evidence does not show that Phrase wins every language pair, narrative voice or game genre.
The useful production loop is source game text to Phrase, contextual translation and review, then controlled delivery back through files, CLI, REST or the official Unity integration. Fonts, variables, plural rules, layout, character voice and cultural fit still need native review and in-game LQA.
HOW YOU USE IT
Use Phrase Strings in the browser, synchronize a repository with the CLI or REST clients, install the official package beside Unity Localization, and connect supported AI clients through the hosted Remote MCP. A Phrase account, suitable paid plan, project configuration and role permission are required for the relevant route.
Language assets and quality gates
Reuse translation memory and terminology, apply AI or professional translation, run documented QA checks, assign review and keep machine output separate from approved release language.
O01O13O14Repository and Unity delivery
Use files, phrase push/pull, REST clients or the official Unity Localization plugin to synchronize reviewed content with a codebase and String Table Collections.
O05O07O10O11O22External-agent entry points
Connect a supported MCP client to the hosted EU or US endpoint, or build bounded tools with REST, CLI and six official SDK families. Existing Phrase roles and selected consent scopes still govern access.
O08O09O11O12O22INTERFACE & EXAMPLES
See how it works
The Phrase Strings editor separates string keys, source and target text, and contextual information into three areas.
Official source retrieved September 2026; interface version not specified.Source: Phrase Support
REVIEW SCOPE
This page evaluates Phrase as continuous-localization infrastructure for game UI, system text, dialogue and live updates from current public evidence. It does not claim hands-on product testing, translation accuracy for a particular language pair, current Unity round-trip reliability, Remote MCP success rates or a measured return on investment.
- Beginner friendlinessSome basics needed
- A translator can enter an already configured job through a structured browser interface, and official CLI and Unity routes shorten common handoffs. A solo developer building the pipeline still needs locales, stable keys, String Tables or file formats, placeholders, metadata, branches, permissions, review states and safe write-back, while the representative annual cash commitment makes unguided trial and error expensive.
- Agent integrationEasy to integrate
- Official hosted EU and US Remote MCP endpoints include account sign-in, revocable consent and View, Create or Modify/Delete scopes, with examples for Claude Code and ChatGPT. REST, CLI and official Ruby, Python, PHP, TypeScript, Java and Go clients provide additional supported routes, so an external agent can work with bounded localization data without relying on browser copy and paste.
Scope, integration requirements and sourcesChecked
Beginner scope: A solo developer or small team structures a representative set of game UI, system and dialogue strings for two target languages and completes one reviewed round trip; not autonomous localization of a full release.
Agent scope: Let Codex, Claude Code, ChatGPT, CI or another external orchestrator inspect a Phrase project, audit terminology or placeholders, create bounded work and propose controlled string changes before human language and build approval.
Confidence: learning Medium · Agent High
- Phrase hosted Remote MCPOfficial
- Phrase Strings REST APIOfficial
- Phrase Strings CLIOfficial
- Phrase Strings API clientsOfficial
- Access
- Hosted MCP uses a current Phrase account and the EU endpoint https://mcp.eu.phrase.com or US endpoint https://mcp.us.phrase.com. Select View, Create or Modify/Delete consent, while the user's existing organization role remains the ceiling. REST, CLI and SDK routes use scoped credentials that must stay in a trusted local or server environment.
- Billing
- The representative Software UI/UX plan is $525 per month billed annually, or $6,300 per year, and includes API access, five Strings projects and listed capacity. Managed words, MTUs, AIUs, professional translation, capacity top-ups and the calling agent are separate limits or cost layers; exact top-up, quarterly and tax totals remain unknown.
- Getting results
- Begin with View consent, list the target project, locales, terminology and placeholder risks, and produce a proposed change set. Add Create for one bounded project, write to a review branch or unapproved state, preview the diff, then require a human language reviewer and game-build gate before delivery. Reserve Modify/Delete for backed-up, auditable and explicitly confirmed work.
- Limitations
- Easy means supported external composability, not autonomous localization or proven current MCP reliability. The caller still owns role design, consent, secrets, pagination, rate limits, retries, destructive-write safety, branch and locale mapping, native review and in-game LQA; recent independent MCP usage evidence was not located.
These are independent editorial judgments, not an overall score or a claim of hands-on agent integration testing.
- Phrase game localization workflow · Official · Learning evidence
- Phrase Unity Strings integration · Official · Learning evidence
- Phrase beginner and workflow records on G2 · Independent · Learning evidence
- Phrase integration records on Capterra · Independent · Learning evidence
- Phrase pricing and plan capacity · Official · Learning evidence
- Phrase Strings CLI · Official · Learning evidence · Integration evidence
- Phrase hosted Remote MCP · Official · Integration evidence
- Phrase MCP starter prompts · Official · Integration evidence
- Phrase Strings API authentication · Official · Integration evidence
- Phrase Strings API pagination and rate limits · Official · Integration evidence
- Phrase Strings API clients · Official · Integration evidence
- Phrase CLI multi-source cleanup issue · Independent · Integration evidence
- Phrase Terms Version 4.2 · Official
- Phrase AI and machine learning statement · Official
KEY FINDINGS
What the evidence supports—and what it does not.
Editorial inferenceMedium confidence
Coordination is the core value
The consistent product value is keeping changing source text, language state, context, terminology, people, review and delivery connected. That matters most after spreadsheets and direct translation calls begin creating version and ownership mistakes.
O01O18O19O20U01U03U04Editorial inferenceMedium confidence
LLMs are being governed, not simply added
Next GenMT and AI Translation Agent are designed to use translation memory, terminology, style and context inside the same workflow. Official descriptions support that direction, but not a universal quality advantage over general-purpose models.
O13O14O15Official factMedium confidence
Unity integration is real and configuration-sensitive
The official package synchronizes Phrase Strings with Unity Localization String Tables, but missing target locales can let a push create work without propagating downstream or showing an obvious error. The Asset Store has too few ratings for a current reliability judgment.
O05O06O07Official factHigh confidence
External-agent access is unusually complete
Hosted regional MCP endpoints, consent scopes, REST, CLI and six official SDKs provide several supported ways for an external agent to inspect or change bounded localization data. This supports an easy integration grade, not hands-free language approval.
O08O09O10O11O12O22Official factHigh confidence
The practical game entry point is $6,300 per year
Software UI/UX lists at $525 per month billed annually and includes Phrase Strings capacity. The $27 Freelancer row has no Strings seat, so presenting it as a game-localization starting price would be misleading.
O02Editorial inferenceMedium confidence
Current game-team evidence remains thin
Phrase-published cases establish use by game companies, while eight independent records expose learning, interface, price and automation risks. They do not establish current Unity reliability, AI dialogue quality, first-pass acceptance or solo-project ROI.
O18O19O20U01U02U03U04U05U06U07U08EDITORIAL VERDICT
Our take
Medium confidencePhrase is a credible shortlist option when a game has changing text, several target languages, multiple contributors and a real need to connect source control, language assets, review and release. Its strongest value is coordination and governance, not one-click AI translation. The $6,300 annual cash entry for the representative game plan, configuration work and human QA make it a poor default for a small one-off prototype. We list it without an Editorial Recommendation until current game-team evidence, Unity experience and accepted-release economics are stronger.
Conclusion scopeMedium-confidence public-evidence judgment for continuous localization and external composability. It does not rank translation quality, current Unity reliability or value against every TMS, direct API or general-purpose LLM.
BETTER FIT
Worth auditioning when
- A live game, DLC or seasonal project whose source text keeps changing after translation begins.
- A team coordinating several languages, branches, reviewers, outside translators or language vendors.
- A Unity Localization project that needs controlled String Table synchronization and can maintain its locale configuration.
- A localization pipeline that benefits from scoped MCP, REST, CLI or SDK access for Codex, Claude Code, CI or internal automation.
POORER FIT
Do not depend on it yet when
- A short prototype with little stable text, one or two target languages and no recurring release cycle.
- A buyer who cannot justify the representative plan's $6,300 annual cash commitment before translation and review.
- A team expecting AI or a vendor order to replace native editing, cultural review, layout checks and in-game LQA.
- Highly confidential or sensitive content uploaded before contract, DPA, subprocessor, region, retention and AI-use review.
WORKFLOW FIT
Where it fits in a production workflow.
Model the source content first
Define stable keys, source and target locales, variables, plural rules, screenshots, character limits, owners and review states before importing a representative slice.
GuardrailBack up existing files and keep sensitive story material out until data and contract terms are approved.
Create drafts inside language controls
Reuse approved translation memory and terms, apply Next GenMT, AI Translation Agent or a selected supplier where justified, then route every target language to a qualified reviewer.
GuardrailMachine fluency is not approval; mark machine output and block it from the release branch until review and QA pass.
Deliver through one controlled route
Choose files plus CLI, REST, or the Unity plugin for the first round trip. Compare stable IDs, locales, variables and statuses before merging the returned content.
GuardrailIn Unity, confirm target locales and propagation explicitly, then inspect fonts, layout, fallbacks and plural behavior in the real build.
Give external agents the smallest useful scope
Start MCP or API automation with View access, list intended changes, add Create for one project, and require a diff or preview before any bulk modification.
GuardrailEnable Modify/Delete only with a backup, branch, bounded project scope, human confirmation, audit trail and tested rollback.
RECOMMENDED WORKFLOW
Use this tool inside a complete production workflow.
Start from a testable brief, then move through prototyping, assets, audio, testing, and release with an explicit handoff and human check at every step.
SHORTEST RESPONSIBLE PATH
Prove one changing-string round trip before committing a game to the platform
This is a low-risk starting workflow synthesized from the evidence.
- 01
Choose 100–200 non-sensitive UI, system and dialogue strings with variables, length limits and two target languages.
- 02
Define stable keys, locales, screenshots or context, terminology, owners and approval states before import.
- 03
Use the fourteen-day trial to import the slice and calculate managed-word use; do not infer the $27 Freelancer plan includes Strings.
- 04
Run one round trip through either CLI or the Unity plugin, deliberately change source text and verify IDs, target locales, variables and status propagation.
- 05
Connect the hosted MCP with View scope, ask it to audit projects, terminology and placeholders, then preview a bounded Create or update task.
- 06
Require native review and in-game LQA, record setup, correction and coordination time, and compare total annual cost with accepted language releases before buying or migrating more content.
PRICING & RIGHTS
The annual platform fee buys workflow capacity—not an accepted localized release
Pricing and terms last checked: Sep 15, 2026
- Software UI/UX is $525 per month billed annually, or $6,300 cash per year. It lists fifteen Strings seats, five Strings projects, unlimited languages, one million managed words, 500,000 TMS processed words per year, one million MTUs, 10,000 AIUs and one Orchestrator workflow.
- Freelancer is $27 per month billed annually but has one TMS seat and zero Strings seats. It is not the public starting plan for managing game strings.
- The trial is fourteen days, includes Enterprise features except paid integrations and does not automatically charge afterward, according to the current pricing page.
- Managed words equal stored source words multiplied by every configured language including the source. Japanese, Korean, Simplified Chinese and Traditional Chinese source characters are divided by two for source-word counting.
- MTUs measure input Unicode code units sent to Phrase Language AI and are separate from managed-word capacity. AI Translation Agent also uses AIUs; neither unit publishes a stable conversion to accepted game-language releases.
- Professional translation ordered in the platform starts at $0.06 per target word. It is optional, supplier- and language-dependent, and is not a quote for a finished game localization.
COST TO OUTPUT
What the representative game plan and optional vendor floor can cover
The platform component uses the public Software UI/UX annual price and capacity. The three translation scenarios assume five target languages, target words equal source words in each language, and the published $0.06 starting rate. Machine translation and professional translation are not both required.
O02O03O04This buys a year of platform access and listed capacity. It does not buy a translated game or guarantee that a small project will use the capacity efficiently.
60,000 managed words, or 6% of the plan, plus a $3,000 professional-translation minimum under the stated assumptions.
300,000 managed words, or 30% of the plan, plus a $15,000 professional-translation minimum under the stated assumptions.
600,000 managed words, or 60% of the plan, plus a $30,000 professional-translation minimum under the stated assumptions.
Public evidence does not supply language mix, translation-memory reuse, supplier quotes, correction time, acceptance rates or in-game LQA needed to calculate this cost.
Actual cost per accepted language release = platform, capacity top-ups, translation or MT, engineering, migration, project management, native editing, rework, layout and in-game LQA divided by language builds that pass the project's release bar. Tax, quarterly terms and top-up prices are also unknown.
Commercial-use condition
Commercial use is allowed with conditions. Version 4.2 says customers retain Customer Content ownership and must hold the rights needed for submitted text, screenshots, translation memories and other assets. Machine and supplier output is not warranted accurate, complete or fit for purpose. Review the order form, DPA, providers and the Terms effective for the account; this page is not legal advice.
Version 4.2 permits some non-text model training and internal research use of Customer Content, while text-generative model training needs prior approval. Phrase's specific Next GenMT and AI Translation Agent statement says those features and their LLM providers do not train on or retain Customer Content. Treat both layers as material, prohibit special-category or sensitive personal information, confirm region, subprocessors, retention and deletion, and recheck Version 4.4 after October 4, 2026. Ordinary subscription cancellation requires at least thirty days' notice under Version 4.2.
PRODUCTION RISKS
Resolve these before production use.
A small project can pay for mostly unused capacity
A 10,000-source-word game with five targets uses only 60,000 of one million managed words. Measure recurring coordination and version-error savings before committing $6,300 annually.
O02O03Unity can look successful while propagation stops
Missing target locales can allow a push or job to occur without downstream integration propagation or a visible error. Add explicit locale and build checks around every synchronization.
O05O06Broad write scopes can damage localization state
MCP offers Modify/Delete, the CLI can push and clean up, and a historical multi-source issue deleted needed keys. Begin read-only, preview bulk work, use backups and branches, and keep a human approval gate.
O08O10O21U07Legal and data handling need a near-term recheck
The research-date Terms and the product-specific AI statement use different levels of scope, and Version 4.4 takes effect soon. Do not migrate unreleased scripts or player-related data on a summary alone.
O15O16O17AI and supplier output still needs release acceptance
Phrase can organize context, terminology, QA and review, but it cannot establish character voice, cultural fit, variable safety, fonts, layout and gameplay meaning without qualified language and game-build checks.
O01O13O14O16O18U01U02NOT VERIFIED
Claims this page does not make
- Current Unity plugin installation, round-trip reliability, large-project performance, failure recovery and time saved in a representative game.
- Remote MCP tool coverage, latency, error rate and independent stability with current Claude Code, ChatGPT or other clients.
- Phrase AI translation quality versus direct DeepL, general-purpose LLMs or human workflows for particular language pairs, UI text and character dialogue.
- Quarterly cash price, taxes, managed-word or MTU top-ups, supplier quotes and the exact checkout total for a specific organization.
- Cost per accepted localized release after translation-memory reuse, engineering, native editing, rework, layout and in-game LQA.
- The practical effect of Terms Version 4.4 after October 4, 2026 and the final account-level order form, DPA, data region, subprocessors, retention and deletion terms.
RESEARCH METHOD
Research-reviewed from public evidence
Research-reviewed from public evidence. We checked twenty-two official product, pricing, capacity, Unity, developer, AI, legal, changelog and game-case sources, then coded eight principal external records across G2, Capterra, Trustpilot, GitHub Issues and Stack Overflow. No platform contributes more than two of eight. Vendor game cases establish workflow adoption only and are not counted as independent satisfaction evidence. We also searched Unity Discussions, the Unity Asset Store, Reddit game communities and GameDev.net without padding the sample with generic localization comments. No registration, subscription, string upload, Unity installation, MCP connection or API call was made or required.
Research windowCurrent product, pricing, integration and legal material checked September 15, 2026; eight principal external records span 2020–2025. Terms Version 4.4 was checked before its October 4, 2026 effective date and requires a post-effective-date recheck.
OFFICIAL SOURCES22
- O01Phrase game localizationChecked Sep 15, 2026
Official game workflow across Phrase Strings, context, translation memory, terminology, QA, suppliers and continuous delivery. Customer examples and outcome claims remain vendor evidence.
- O02Phrase pricing and plan capacityChecked Sep 15, 2026
Current annual-billing prices, seats, projects, managed words, TMS words, MTUs, AIUs, trial and professional-translation starting rate.
- O03Cost-effective word management in Phrase StringsChecked Sep 15, 2026
Defines managed words as stored source words multiplied by all configured languages, including the source language; this is capacity, not annual translation output.
- O04MTU consumption and managementChecked Sep 15, 2026
Defines MTU input-code-unit billing, native versus ecosystem engine rates and repeated workflow consumption. MTUs are separate from managed-word capacity.
- O05Unity integration for Phrase StringsChecked Sep 15, 2026
Official bidirectional workflow for Unity Localization String Table Collections. It also documents a target-locale configuration that can stop downstream propagation without a visible error.
- O06Phrase Unity Asset Store listingChecked Sep 15, 2026
Unity lists the package as a Verified Solution, but the listing did not have enough ratings to support a current user-experience judgment.
- O07Official Phrase Unity plugin repositoryChecked Sep 15, 2026
Official source repository for the Unity integration. Repository existence does not establish round-trip reliability in a particular project.
- O08Phrase hosted MCP serverChecked Sep 15, 2026
Documents EU and US Remote MCP endpoints, account sign-in, consent scopes and supported clients including Claude Code and ChatGPT.
- O09Phrase MCP starter promptsChecked Sep 15, 2026
Vendor examples for project inspection, terminology, placeholders, jobs, notifications and bulk updates. They are workflow ideas, not independent success evidence.
- O10Using the Phrase Strings CLIChecked Sep 15, 2026
Official phrase init, push, pull and API-endpoint CLI route for repositories and CI.
- O11Phrase Strings API authenticationChecked Sep 15, 2026
Official token and JWT authentication for the Strings API. Credentials and least-privilege access remain caller responsibilities.
- O12Phrase Strings API pagination and rate limitsChecked Sep 15, 2026
Documents pagination, concurrency and request-rate boundaries that an automation wrapper must handle.
- O13Phrase AI Translation AgentChecked Sep 15, 2026
Describes agentic translation using Phrase language assets and quality controls. Product claims do not prove quality for every language pair or game-writing style.
- O14Phrase Next GenMTChecked Sep 15, 2026
Current Phrase machine-translation direction using context and language assets. It is not an independent comparison with general-purpose LLMs.
- O15Phrase AI and machine learning statementChecked Sep 15, 2026
Says Next GenMT and AI Translation Agent do not train or fine-tune on Customer Content and that LLM-provider agreements prohibit retention and training; broader Terms still need to be read alongside it.
- O16Phrase Terms Version 4.2Checked Sep 15, 2026
Terms current on September 15, 2026: Customer Content ownership and input duties, some ML and internal-research rights, accuracy disclaimers, sensitive-data restrictions and thirty-day cancellation notice. Not legal advice.
- O17Phrase legal terms Version 4.4Checked Sep 15, 2026
Upcoming Terms published before review and scheduled to take effect October 4, 2026. They trigger a short-term recheck rather than replacing Version 4.2 early.
- O18Phrase-published GameHouse caseChecked Sep 15, 2026
Vendor case showing a game company managing nearly twenty languages while retaining human translators. It establishes a possible workflow, not general savings or satisfaction.
- O19Phrase-published MIXI caseChecked Sep 15, 2026
Vendor case covering four-language game UI, system and marketing content. Selection bias prevents treating it as independent quality evidence.
- O20Phrase-published 37GAMES caseChecked Sep 15, 2026
Vendor case for a large game-localization pipeline. Its scale and reported outcomes cannot be projected onto a solo project.
- O21Phrase CLI changelogChecked Sep 15, 2026
Official history showing later multi-source cleanup support after a destructive historical edge case. Version-lock and preview write operations.
- O22Official Phrase Strings API clientsChecked Sep 15, 2026
Official Ruby, Python, PHP, TypeScript, Java and Go clients for the Strings API.
INDEPENDENT EVIDENCE RECORDS8
- U01Antti M. Phrase review on G2Checked Sep 15, 2026
A 2023 small-business Product Owner reports useful translator invitations, glossary management and correct-format downloads, alongside beginner complexity and room for better screenshot context. Seller-invited and incentivized.
- U02Validated localization user Phrase review on G2Checked Sep 15, 2026
A 2025 user reports an adaptable but complex interface, renewal, quota and workflow friction. The record predates some 2026 AI and MCP changes and is incentivized.
- U03Mateusz K. Phrase review on CapterraChecked Sep 15, 2026
A 2020 product-marketing account values cross-department work, history and screenshots while noting weak immediate visibility for duplicate-string recency. Historical and non-game evidence.
- U04Marco F. Phrase review on CapterraChecked Sep 15, 2026
A 2020 engineering account reports integration across website, email and application content, with some language navigation friction. It supports long-term integration feasibility, not current game quality.
- U05Phrase workflow-interruption review on TrustpilotChecked Sep 15, 2026
A 2025 user says pop-ups and notifications interrupted a previously preferred CAT workflow. A single unverified report and not a game benchmark.
- U06Phrase pricing and quota review on TrustpilotChecked Sep 15, 2026
A 2025 user reports pricing, quota, support, account-lock and credit-calculation friction. Trustpilot warns that its small Phrase set may not be representative.
- U07Phrase CLI multi-source cleanup issue #147Checked Sep 15, 2026
A 2024 issue reports phrase push --wait --cleanup deleting keys needed by another source in a multi-source project. Later changelog support means this is a historical safety lesson, not a claim that current versions remain broken.
- U08Phrase and Xcode String Catalog questionChecked Sep 15, 2026
A 2024 iOS developer sought a conversion and upload path because Phrase did not then support Xcode String Catalogs. Historical format-gap evidence, not a current Unity result.
DISCLOSURE
How this research was supported
Independent editorial research. MakeGameWithAI has no affiliate link, sponsorship, vendor-provided account, trial extension, credits, translation order, interview or technical support for this review. Owner Review approved the continuous-localization positioning, some_basics / medium beginner decision, easy / high external-Agent decision, representative-plan and managed-word framing, commercial and data boundaries, and no Editorial Recommendation on September 15, 2026.
This public research review will be revisited when pricing, terms, product versions, or material new evidence changes.