Cursor Translate

Serve your Cyrillic markdown docs to Cursor agents as cached English translations — and cut the context tokens agents burn re-reading them.

Russian prose tokenizes ~1.8–2× worse than English, and agent workflows read the same docs over and over: project wikis, ROADMAPs, skills, reports. cursor-translate translates a doc once per version on a cheap nano tier (agent --print --mode ask --model gpt-5.4-nano-none, billed to your existing Cursor subscription — no separate API key) and serves the cached English on every subsequent Read — in the IDE, the agent CLI, and (via committed caches or MCP) Cloud Agents.

Sibling project claude-translate builds on the same engine (@cursor-translate/core) for Claude Code.

The core value — and its honest economics

The saving mechanism is one and automatic: the plugin's preToolUse hook redirects every Read of a Cyrillic .md/.mdx to a cached English translation — in every session, every subagent, every project.

Translation is an investment (one nano spend per doc version), serving is the return (every read). Which means:

• Pays off: stable docs that agents read often — knowledge bases, project registries, ROADMAPs, rules and skills.
• First-time translation is the investment: a fresh doc repays itself in ~1–3 reads (check the break-even reads line in your own report).
• Edits are cheap: re-translation is section-incremental (cache.incremental: section, default) — only the changed ##/### sections are re-billed. And since agents Read a doc around every Edit, reads ≥ edits in practice, and a single read of a mid-size doc typically covers a one-section update. A doc only stays net-negative when it's bulk-rewritten often and rarely read afterwards (one-off scratch notes).
• Saves ~nothing: code-heavy sessions that rarely Read Cyrillic markdown — the savings scale directly with how much Cyrillic documentation your agents actually read.

Don't take the pitch's word for it — every cache hit, every translate spend and every missed opportunity is logged. Pull your own numbers any time:

cursor-translate report --days 7

ROI operational (docs + CLI/MCP prompts — excludes warmup):
  doc cache served (lazy read / MCP resolve_doc): ~… tokens saved (N reads)
  incremental doc translate: ~… tokens
ROI investment (one-time doc cache warmup):
  break-even reads (warmup ÷ avg savings/read): ~N more doc_cache_served events
session opportunity (interactive — not auto-translated):
  user_prompt (RU sent as-is to main model): ~… tokens (N events)

How to read it: doc cache served is your realized saving; the doc translate lines are what the nano tier spent to earn it; session opportunity is what the extras below could additionally capture (and what they'd cost you in trade-offs).

Installation

npm install -g cursor-translate @cursor-translate/mcp
cursor-translate init --path
source ~/.zshrc   # or open a new terminal

Enable the plugin in Cursor:

ln -sf "$(npm root -g)/cursor-translate/plugin" ~/.cursor/plugins/local/cursor-translate

Restart Cursor → enable cursor-translate in settings. MCP tools translate and resolve_doc activate via plugin/mcp.json after init --path.

Without global install:

npx cursor-translate init --path

Requires the agent CLI logged in (Cursor subscription) and Node ≥ 24.

Quick start

cd ~/Projects/your-repo
cursor-translate docs --dry-run   # see what would be cached
cursor-translate docs             # warm the EN cache (one-time nano spend)
cursor-translate report --days 7  # savings vs costs (full economy ROI)

How the core works

Lazy translate on Read

The plugin's preToolUse hook (matcher Read): if the file is .md/.mdx with Cyrillic and the cache is missing or stale (sha mismatch), it translates via nano, caches under ~/.cursor/translate-proxy/cache/<project>/…en.md, and rewrites the tool call's path to the cache. It also injects a context note telling the agent to edit the original file, never the cache. Everything fails open: no CLI, quota exhausted, timeout → the original Russian file is read.

Large cold/stale docs: when a file exceeds cache.lazy_read_max_chars (default 50 000) or cache.lazy_read_max_chunks (default 3), lazy translate is deferred — the agent reads Russian and sees a pre-warm hint. Run cursor-translate doc <file> to warm manually.

Incremental cache: cache.incremental: section (default) re-translates only changed ## / ### sections; section payloads live in *.en.sections.json sidecars next to flat *.en.md files served to Read.

Shared cache with claude-translate

Before spending on a translation, the doc cache checks the sibling install — claude-translate keeps the same cache format under ~/.claude/translate-proxy. A fresh entry (sha match against the current source) is copied over as action: sibling_copy with zero translate cost; only if the sibling is also missing or stale does a real translation run. Works in both directions.

Config: cache.share_siblings: true (default). Override or disable: CURSOR_TRANSLATE_SIBLING_HOMES="/path/one:/path/two" (empty string disables).

CLI commands

Command	Purpose
init [--path]	Config, glossary, hook assets, bin wrappers, optional shell PATH
doc <file>	Translate one file → global cache
docs [path]	Scan project *.md with Cyrillic → cache all
resolve <file>	Lazy: ensure EN cache, print readPath
hook-resolve	stdin JSON for the preToolUse Read hook
prompt "<text>"	RU→EN translate to stdout
agent [agent flags] -- "<prompt>"	Full RU→EN → agent -p → EN→RU
report [--days 7]	Metrics by source + ROI break-even
backfill-costs	Backfill translate_cost_usd from agent JSON logs

Metrics sources (~/.cursor/translate-proxy/metrics.jsonl)

source	Trigger
doc_cache_served	Lazy read / MCP served EN cache (realized savings)
doc_translate_cost	Doc translation spend (warmup_translate = batch docs; lazy_translate = on-demand)
prompt_translated / response_back_translated	CLI agent & prompt, MCP translate
user_prompt / agent_response	Opportunity audit from hooks (what auto-translate would save)
file_read / subagent_task / subagent_summary	Read and subagent audits

cursor-translate report --days 7 includes ROI full economy: doc cache savings, incremental break-even reads, translate spend, and session opportunity estimates.

Extras (opt-in, experimental)

Everything below is configured in your local ~/.cursor/translate-proxy/config.yaml — created on your machine by cursor-translate init from the packaged template and yours to edit.

The platform reality that shapes these features: Cursor's beforeSubmitPrompt hook cannot rewrite your prompt (block + message only), afterAgentResponse has no output fields, and — unlike Claude Code's MessageDisplay in the sibling project — Cursor has no display-layer substitution at all. So in the IDE, Russian chat goes to the model as-is and replies are whatever the model writes; prompt- and reply-side translation exists only outside the IDE loop. The doc cache above has no such limitation — which is exactly why it's the core.

Full agent wrapper (headless) — the only place prompts and replies get auto-translated

cursor-translate agent --model composer-2.5 -- "сделай ревью PR и опиши риски"

User RU → nano (translate in)
       → agent -p (your model; hooks/doc-cache still active)
       → nano (translate out, optional — response.back_translate)

Skip back-translate for English output:

cursor-translate agent --no-back-translate -- "explain the auth flow"

For scripts, cron jobs and CI — the RU→EN→agent→RU loop runs outside the model, so the platform limits above don't apply.

MCP tools: translate / resolve_doc — the primary path for Cloud Agents

Cloud Agents don't run user-level hooks — there, resolve_doc is the explicit way to get the EN cache for a Cyrillic doc, and translate is an on-demand RU↔EN capability on the cheap tier (glossary-aware, metered). Add the snippet from templates/agents-md-cursor-translate-snippet.md to your project's AGENTS.md or CURSOR.md (or a ## cursor-translate section) — it tells Cloud Agents and local sessions when to call the tools. In the IDE with the plugin enabled these mostly stay idle — the preToolUse hook already did the job — and that's by design.

Opportunity audits

beforeSubmitPrompt / afterAgentResponse hooks log what auto-translation would have saved (user_prompt, agent_response sources) — that's the session opportunity block in report. Audit only, no behavior change.

Two-tier model strategy

Tier	Model	Used for	~API rate (in/out per 1M)
Main agent	Sonnet / Opus / Composer (your IDE setting)	Code, reasoning, tools	$3–15+ / higher
Translate tier	gpt-5.4-nano-none (default)	RU↔EN prose only	~$0.05 / $0.40

Translate hops run as:

agent --print --mode ask --model gpt-5.4-nano-none -p "<translator prompt>"

Billing draws from your Cursor subscription API usage pool at the model's published rate — no separate OpenAI API key required. See Cursor models & pricing.

When nano quota is exhausted, doc cache falls back to composer-2.5 (translator.doc_fallback_model); prompt/response translation is skipped (fail-open). Quota latch auto-expires after 30 minutes (CURSOR_TRANSLATE_QUOTA_TTL_MIN to override).

Override: CURSOR_TRANSLATE_MODEL=gpt-5.4-nano-low or translator.model in ~/.cursor/translate-proxy/config.yaml. CI/headless without a subscription: CURSOR_TRANSLATE_PROVIDER=openai + OPENAI_API_KEY.

Plugin contents

• Hooks: preToolUse lazy read (600s timeout), beforeSubmitPrompt / afterAgentResponse opportunity audits, sessionStart context note. All guarded by CURSOR_TRANSLATE_HOP=1 against recursion; disabled features exit before booting node.
• Rules: translate.mdc (lazy read policy, glossary), mcp-translate.mdc (when agent must call MCP).
• MCP: translate + resolve_doc via @cursor-translate/mcp (npx @cursor-translate/mcp in plugin/mcp.json).

Config

~/.cursor/translate-proxy/config.yaml — your local file, created by init from templates/config.yaml:

translator:
  provider: cursor-cli
  model: gpt-5.4-nano-none

Custom translation rules: .cursor/cursor-translate.md, a ## cursor-translate section in CURSOR.md/AGENTS.md, or ~/.cursor/translate-proxy/cursor-translate-rules.md. Project glossary: .cursor/cursor-translate-glossary.yaml.

Related docs

• Runtime guide — hook contracts, IDE vs CLI vs Cloud limits, config and env reference, metrics, troubleshooting, fail-open guarantees
• MCP setup — translate and resolve_doc for Cloud Agents
• Cloud Agents playbook — committed EN caches, MCP checklist, IDE hooks gap
• Publishing — npm release flow, CI, lockfile notes
• Cursor Marketplace — plugin submission checklist
• Changelog

License

MIT