LensWordLearn words that stick.

Spaced repetition, forced recall, and memory-palace visualization — across a web app, desktop shell, browser extension, and MCP server.

Setup (5 min tutorial)

Choose your surface

View on GitHub

🚀

Setup

A hands-on tutorial: run LensWord end to end with Docker Compose and see it work in minutes. Start here if you're new.

Start the tutorial

🧩

Install

Task-oriented how-to guides for every surface — web, desktop, browser extension, MCP server, self-hosting, local AI — plus troubleshooting.

Browse install guides

🏛️

Learn

Understand how LensWord is put together and which surface fits your goal — architecture, design decisions, and the surface comparison.

Read the explanations

📚

Reference

Technical lookup material — ADRs, config/env vars, the changelog, release process, and what's actually been verified.

Browse reference

🔍

Trust & verification

What's been tested and how, what hasn't, and every known gap — stated plainly.

See what's verified

🤝

Contributing & support

Development setup, the pull request process, and how to sponsor the project.

Contribute

What LensWord is

LensWord is an open-source vocabulary trainer that forces spaced-repetition recall instead of passive review, and lets you anchor words spatially in a memory palace (the method of loci). It's a FastAPI + Postgres backend behind a Vite/React web app, with an optional desktop shell, browser extension, and MCP server for AI clients — see Choose your surface for what each one is good for and how mature it is today.

Surface	Purpose	Requires	Status
Web Application	Browser-based vocabulary learning and review interface.	apps/backend	public
Desktop Application	Tauri 2 desktop shell wrapping the web frontend, remote-only (no bundled backend or database).	apps/backend (remote, reachable server required)	unreleased No git tag or GitHub Release has ever been created. README states explicitly: 'No release has been published yet.'
Browser Extension	Chrome MV3 extension to capture selected words from any webpage into a LensWord group.	apps/backend	public-dev-mode-only Functional but not published to the Chrome Web Store; zero CI coverage. v1 scope hardcoded to Spanish, no translations.
MCP Server	Model Context Protocol server exposing LensWord actions to MCP clients (Claude, Codex, Cursor, etc.) over stdio by default, or Streamable HTTP when explicitly enabled.	apps/backend	public-source-install-only Not published to PyPI. No CI coverage despite having its own pytest suite. Remote HTTP transport is double-gated off by default (REMOTE_MCP_ENABLED on the backend + LENSWORD_MCP_TRANSPORT/LENSWORD_MCP_REMOTE_ENABLED on the MCP process) and has no live interop test against a real third-party MCP host.
Local CLI	Command-line entry point ('lensword') distinct from the MCP server, with add/explain/diagnose/review/import-context subcommands.	apps/backend (all subcommands except import-context)	public-source-install-only import-context is read-only/offline by design (never writes, never contacts the backend, redacts secrets, 50,000-char default cap). Other subcommands need the same backend env vars as the MCP server.

How these docs are organized

This site follows the Diátaxis framework, so each page has one job:

Setup — a tutorial. Follow it once, end to end, to get LensWord running.
Install — how-to guides. Task-oriented recipes for a specific surface or problem.
Learn — explanation. Understand how the pieces fit together and why.
Reference — lookup material. ADRs, config, the changelog, and what's verified.

No tagged release exists yet for any surface — this documentation describes the current development branch, evaluated with evidence rather than assumed. See Trust & verification for the full picture.