Locale-correctness tools for AI agents. Agents are systematically wrong about
everyday non-US data — they read 03/04 as March 4 in the UK, apply today's VAT
rate to a historical invoice, format €1,234.50 for a German reader, and miss
bank holidays in delivery dates. localecheck gives them verified answers and
flags impossible inputs instead of guessing.
Coverage: UK + USA (v1), expanding. Free, no API key.
| Package | Install | Use | |
|---|---|---|---|
| TypeScript (primary) | packages/localecheck-ts |
npx -y localecheck (live on npm) |
MCP server and importable library |
| Python | packages/localecheck-python |
from source (PyPI release coming soon) | MCP server, importable library, REST API |
Both wrap the same authoritative libraries (libphonenumber, CLDR/Intl, date-holidays) and share the same hand-curated tax data.
parse_date · validate_phone · format_currency · tax_rate (VAT/sales-tax
by date) · is_holiday / next_holiday · parse_address
{ "mcpServers": { "localecheck": { "command": "npx", "args": ["-y", "localecheck"] } } }Then ask naturally — the agent picks the tool: "A UK invoice dated 1 June 2009 — what VAT rate?" → 15%.
packages/localecheck-ts— primary TypeScript package (MCP + library)packages/localecheck-python— Python package + FastAPI REST APIeval/— benchmark harness: how often raw LLMs get this wrongbenchmark.html— shareable results page ·BUSINESS_MODEL.md— strategy ·PUBLISH.md— release steps ·LOCAL_TEST.md— test before publishing
UK VAT history is reliable for the modern era (standard rate). US state sales-tax values are 2025 state base rates (local taxes add on top) — verify against an authoritative feed before live invoicing. Addresses parse + extract a valid postcode/ZIP but don't verify deliverability.
Code: Apache-2.0. Curated tax data: see BUSINESS_MODEL.md.