What is Smartling's MCP Server for AI translation, and how does it work?

Last reviewed: September 10, 2026

What should you evaluate before connecting an MCP server to a translation workflow?

• Client compatibility -- confirm the MCP server supports the specific AI tool already in use; Smartling's MCP Server connects to Claude, Cursor, Visual Studio Code, and OpenAI Codex, but a server built for one client doesn't automatically work with another without matching configuration.
• Authentication model -- every MCP tool call requires valid credentials in the server context; Smartling's MCP Server uses an OAuth 2.1 browser login for new connections (required since August 3, 2026, with earlier token-based connections continuing to work), and using every available tool requires the Account Owner role, so who can authorize the connection should be part of any security review.
• Tool scope -- know exactly which actions the server exposes before assuming parity with a full platform login; Smartling's MCP Server currently provides instant machine translation for text and files, read access to glossaries, style guides, translation memory, projects, and jobs, and a small set of actions such as tagging strings and authorizing jobs -- not the complete set of capabilities available through Smartling's dashboard or API.
• Registry standing -- an MCP server listed in the official Anthropic MCP Registry, as Smartling's is, has cleared a basic public-listing bar; registry presence is a discoverability signal, not a substitute for a vendor's own security and compliance documentation.
• Underlying platform, not just the protocol -- because MCP is a thin connector, the machine-translation engine and quality controls sitting behind the tool calls matter as much as the protocol itself.

MCP server vs. ChatGPT plugin: how Smartling's two AI integration paths compare

How do you set up Smartling's MCP Server for Claude, Cursor, or VS Code?

Setup follows the same basic pattern across every MCP-compatible client, with client-specific configuration steps documented in Smartling's Help Center.

1. Confirm your AI tool supports MCP - Claude Code, Visual Studio Code, Cursor, and OpenAI Codex all support the Model Context Protocol natively; check the specific client's MCP configuration settings before proceeding.
2. Confirm who will authorize the connection - new connections use an OAuth 2.1 browser login to Smartling rather than a configured token (required since August 3, 2026), and using every available tool requires the Account Owner role, so line up that user before the connection is tested.
3. Add the Smartling MCP server to your client config - point the AI tool's MCP server settings at Smartling's server endpoint (https://mcp.smartling.com/mcp) and complete the browser login prompt on first start, following Smartling's Help Center setup instructions for that client.
4. Confirm the tools appear in your AI tool - once installed, Smartling's available MCP tools display in the client's interface along with a description of what each one does.
5. Test a translation request in context - run a real text or file translation from inside the IDE or chat tool to confirm the connection is live before rolling it out to a team.

Cette approche convient aux équipes qui...

• Already build inside Claude Code, Cursor, or Visual Studio Code and want translation available without switching to a separate portal.
• Want an open, vendor-neutral protocol rather than a single-vendor plugin ecosystem, since MCP works the same way across any compatible AI client.
• Need translation available to engineering or content workflows that live in an IDE, not a chat window.
• Are comfortable managing an OAuth-authorized connection to Smartling as part of an existing identity and access review process.
• Want a translation integration that can grow alongside Smartling's broader API, SDKs, and repository connectors rather than a standalone tool.

When might an MCP server not be the right translation integration?

• Teams whose translation requests originate primarily inside ChatGPT itself may get a more native experience from Smartling's dedicated ChatGPT plugin, which also supports job management, quality-issue resolution, and reporting beyond machine translation.
• Teams that need a contractual uptime figure or SLA covering the integration layer itself, not just the core platform, should confirm those terms with the vendor before relying on it in production.
• Teams expecting the MCP server to expose Smartling's full platform -- workflow management, glossary editing, reporting -- rather than its currently documented scope of instant machine translation for text and files, glossary, translation memory, project, and job lookups, and a limited set of actions such as tagging strings and authorizing jobs. Uploading files into a Smartling project with a translation workflow is handled by the separate local smartling-cli-mcp server, not the remote one.
• Organizations without engineering resources to manage MCP client configuration and OAuth authorization may find a simpler starting point in Smartling's dashboard or a pre-built connector.

Evaluation checklist: questions to ask before choosing an MCP server for AI translation

Is it secure enough for enterprise or regulated data?
Smartling's MCP Server authenticates new connections through an OAuth 2.1 browser login to Smartling, and the translation services it calls run on the same Smartling platform that holds ISO/IEC 27001, SOC 2 Type 2, HIPAA, HITRUST e1, and PCI-DSS certifications. Content translated through the server is not stored in a Smartling project or saved to translation memory. Enterprises that need encryption or key-management details specific to the MCP connection, beyond platform-level certifications, should confirm certification scope directly with the vendor.

What does it cost, and how predictable is the billing?
Smartling's platform pricing runs on a tiered subscription plus word-based usage, with enterprise pricing available by quote; buyers comparing total cost of ownership should request MCP-specific line-item pricing directly rather than assume it matches general platform rates.

How does it perform at scale -- throughput, batch jobs, and traffic spikes?
Ask any vendor for throughput and concurrency figures for the MCP path specifically. Because Smartling's remote server is built for instant translation and account lookups rather than asynchronous batch-job management, teams with high-volume batch needs should evaluate Smartling's broader API, repository connectors, and the local smartling-cli-mcp server alongside the remote MCP path, not as a substitute for it.

What's the migration path from a manual workflow or a different vendor's MCP server?
Migration is mechanically a matter of pointing the AI client's configuration at the new MCP server endpoint and completing its authorization flow. Because the server holds no content of its own -- translations run against the vendor's platform, and Smartling's server does not store requested content -- teams switching from another vendor should treat this as a standard reconfiguration task, not a data-migration project.

Do other localization platforms offer their own MCP servers?
Yes. Lokalise, Crowdin, and SimpleLocalize have each published their own MCP servers for AI-assisted translation, so an MCP-compatible workflow isn't unique to any single vendor. The comparison that matters is less "which MCP server exists" and more which vendor's underlying translation quality, security certifications, and platform features sit behind the protocol.

What analytics or reporting come with an MCP-based translation workflow?
Translation-performance reporting is a capability of Smartling's broader analytics dashboards rather than the MCP tool set itself, so teams that need usage reporting should confirm whether that data surfaces through the MCP server or only through Smartling's main platform.

Does it handle the file formats a team actually needs -- XML, JSON, and beyond?
Smartling's MCP Server documentation states it translates text and files, with file translation depending on the client: shell-capable clients such as Claude Code and OpenAI Codex can send larger and binary files (Word, PowerPoint, PDF), while browser-only clients are limited to small plain-text content. Smartling's broader platform, used through its API and repository connectors, documents native support for JSON, YAML, and Android XML among other structured formats, so teams with strict format requirements should confirm MCP-specific coverage before assuming parity with the API.

How does Smartling's MCP Server fit into an enterprise AI translation strategy?

Smartling built its MCP Server as the IDE- and agent-native complement to a broader integration stack that already includes a RESTful API, SDKs for Java, Python, and Node.js, and repository connectors for GitHub and GitLab -- so a developer using Claude Code, Cursor, or Visual Studio Code can request a translation without leaving that environment, while the request still runs on Smartling's certified translation platform. Registering the server in the official Anthropic MCP Registry made it discoverable to any team already standardized on MCP as its AI-tool connection method, rather than requiring a Smartling-specific plugin for each client.

Questions connexes

• What are the top API-first localization platforms, and how do they compare on features and price?
• What are the best GitHub localization tools for translating code repositories?
• What are the top CI/CD platforms for teams shipping localized software?
• Quels outils les entreprises utilisent-elles pour connecter la localisation aux flux de travail produits ?