sandy, created by James Telfer, is an MCP server for AI-text-localization that integrates language models into developer pipelines. It automates translation and in-context text adjustments so localization steps execute from the command line. Key elements include a Bun-powered runtime, native Claude Code plugin integration, and CLI tooling. The tool targets software developers, localization engineers, and AI-focused teams who want programmatic control of multilingual text handling.

Automates developer-facing localization actions inside terminal workflows

sandy acts as a bridge between LLMs and code-centric localization pipelines, exposing localization operations as an MCP-compatible server and a CLI. The tool registers with Claude Code as a plugin, enabling invoked agents to run translation and text-processing routines without switching to external GUIs. That makes it suitable for teams that prefer scripted or terminal-based workflows and require repeatable automation for routine text edits.

Produces agent-ready context rather than polished, human-vetted translations

Outputs are intended as AI-assisted text transformations for downstream review, since the tool supplies context and automation hooks to a language model rather than a final human-verified localization asset. Users should treat generated text as model-produced suggestions and apply human validation for critical content, particularly for technical, legal, or customer-facing strings.

Requires specific runtime and environment conditions to run predictably

Operational constraints include a Bun runtime and MCP-compatible environments; sandy requires Bun 1.3 or newer and targets platforms that support MCP servers. Supported shells include macOS and Linux, and Windows through WSL or compatible shells. Installation flows use Bun commands, so the runtime choice and shell environment affect deployment and CI integration strategies.

Integrates into modern developer ecosystems with automated registration and packaging

Setup focuses on developer ergonomics for code repositories and tooling: the project offers an automated registration process for Claude Code and publishes as an npm-scoped package under The developer behind the project is an active contributor in related ecosystems, which supports ongoing maintenance and community-driven extensions.

A practical choice for Claude Code-focused developers who accept its narrow scope

Given its acceptance inside the MCP developer community and its orientation toward command-line automation, sandy suits developers who need programmatic localization hooks in code workflows. Teams that require graphical tools or vendor-agnostic GUI integrations may find its focus limiting. Practical tip: pair generated translations with human review for critical content to reduce risk from model-generated errors.