Glossa is a web-based language learning app that lets you upload your own PDFs and EPUBs and read them in 13 languages with click-to-translate, audio narration, and built-in spaced-repetition vocabulary practice. It turns any book you already want to read into a graded reader for the language you are learning.

What languages does Glossa support?

Glossa supports 13 languages: English, Spanish, French, German, Italian, Portuguese, Dutch, Swedish, Norwegian, Danish, Polish, Russian, and Turkish. Word difficulty is scored from corpus frequency data, so coverage quality tracks how much text a language has in that corpus. Glosses and translations can be shown in 18 languages, those 13 plus Chinese, Japanese, Korean, Arabic, and Hindi. More languages are in preparation.

Can the translations be in my own language instead of English?

Yes. Glosses and sentence translations can be shown in 18 languages: the 13 you can learn plus Chinese, Japanese, Korean, Arabic, and Hindi. Glossa uses your native language by default and you can change it for each book, so a Chinese speaker can read an English novel with Chinese glosses, or a Hindi speaker a Spanish one with Hindi glosses. Chinese, Japanese, Korean, Arabic, and Hindi are available as gloss languages only, not yet as languages to learn.

What file formats can I upload?

PDF and EPUB, up to 100 MB per file. Books upload directly from your browser to Azure Blob Storage via a short-lived signed URL — the application server never sees the raw bytes in transit.

Yes. The free tier includes 2 books and 10 distinct pages of reading, with every feature unlocked (translation, audio narration, vocabulary games). Pro is $9.99 per month billed annually ($119.88 a year), $14.99 month-to-month, or a one-time $4.99 7-day pass, and removes the book and page limits. Annual plans are refundable in full within 30 days.

What is included in Pro?

Unlimited books, unlimited pages, AI-generated chapter summaries, and faster book processing. Pro removes limits rather than gating features — translation, audio narration, and vocabulary practice are also fully available on the free plan.

Does Glossa work on mobile?

Yes, Glossa runs in any modern mobile browser. There is no native iOS or Android app yet; the responsive web app is the only way to use Glossa on phones today.

Is my uploaded content private?

Yes. Uploaded books are stored in a private Azure Blob container accessible only to your account, are never shared with other users, and are never used to train AI models. We do not sell user data.

How does Glossa decide which words to translate?

Glossa estimates how difficult each word in your book is for you, and only the words above your level get a translation written underneath — everything else stays as original text. It adjusts to your level automatically: the more you read and practise, the better it fits you. You can also set it yourself with a slider in the reader — a fuller slider shows more glosses, an emptier one fewer. The slider's stops carry CEFR labels (A1–C2) as a familiar reference point, not an official CEFR assessment.

How does the spaced-repetition vocabulary practice work?

Words you tap while reading are saved to your vocabulary list. Glossa surfaces them in short games — matching, fill-in-the-blank, multiple choice, parts-of-speech tagging, and sentence-scramble — scheduled so you re-encounter each word at increasing intervals based on how confidently you answered last time.

How is Glossa different from Duolingo?

Duolingo teaches with a fixed curriculum of synthetic exercises. Glossa is a reader — you upload real books you actually want to read, and translation and vocabulary practice are built around that content. Duolingo suits absolute beginners; Glossa suits learners who can read elementary sentences and want to engage with longer material.

How is Glossa different from LingQ?

LingQ has a large catalog of curated lessons, a learner community, and native mobile apps, and it can import your own ebooks by splitting them into lessons. Glossa keeps your PDF or EPUB as one book and writes translations between the lines for only the words above your level, adjustable with a slider, where LingQ has you tap words to look them up. Choose LingQ for curated content and a phone app; choose Glossa to read a book you chose with glosses that adapt to you.

How is Glossa different from Readlang?

Readlang is primarily a browser extension for translating words in web articles. Glossa is book-centric — built around uploading PDFs and EPUBs, with audio narration, level-adaptive glossing, and structured vocabulary games. Choose Readlang for web articles; choose Glossa for full books.

Can I import books from Project Gutenberg?

Yes. Glossa has a built-in browser for the Project Gutenberg catalog of public-domain titles. Pick a book in your target language and it imports in seconds without leaving the app.

What voices does Glossa use for audio narration?

Azure Neural Text-to-Speech voices, native-speaker quality in each supported language. Every user gets a default voice per language; additional voices can be unlocked using Glossa Points earned through reading.

Connect an AI assistant to the books you are reading

Live since September 2026. Pro feature, read-only.

Ask an AI assistant what you should read next and it has to guess. It does not know which book you are 40% through, that you read twenty pages yesterday and none for the nine days before, or that apenas has defeated you eleven times. So it answers the way it answers everyone: a list of famous titles and some encouragement.

Glossa ships an MCP connector that closes that gap. Add one URL to Claude, and the assistant can read your Glossa library, your reading history and your saved vocabulary before it says anything. The advice stops being generic because the assistant finally has the facts.

What MCP actually is

The Model Context Protocol is an open standard for letting an AI assistant call out to an external service — a “connector” in Claude’s interface, a “custom connector” or “MCP server” elsewhere. The assistant is handed a short list of things it may ask for, and it asks for them when they are relevant. There is no browser extension, no copying and pasting your stats into a chat window, and no scraping.

Glossa’s connector lives at https://readglossa.com/mcp. It speaks the standard protocol with standard authentication (OAuth 2.1 with dynamic client registration and PKCE), so any client that implements MCP properly can talk to it.

What the assistant can see — all six of them

The connector exposes exactly six tools. That is the whole surface; there is no seventh:

Your learning profile — target language, level, how much of the text you have glossed, streak, plan, lifetime totals.
Your library — the books on your shelf and how far through each one you are, most recently read first.
Your reading activity — pages per day and per book over a recent window.
Your saved vocabulary — words you have saved with their review state, filterable to just the ones due or the ones you keep getting wrong.
Catalog search — the public-domain catalog, filtered hard by language.
Book recommendations — the same ranked recommendations Glossa’s own Browse tab produces, each one carrying the reason it was picked.

What it cannot do

This matters more than the feature list, so it is worth being blunt about. Every tool is read-only. There is no tool that writes anything, and that is enforced on our side, not left to the assistant’s good judgement. An assistant connected to your Glossa account cannot:

upload, import, edit or delete a book;
change your level, your reveal setting or any preference;
mark vocabulary as known, or alter your review schedule;
see your payment details, card or billing information;
change anything about your account, or delete it.

Write tools — importing a book you discussed, nudging a setting — are a plausible future addition, and if they ever land they will be labelled as writes so your client asks you before running them. Anything that touches the account itself is permanently out of scope: a hallucinated deletion is not recoverable, so the tool simply does not exist.

You can disconnect at any time from Account → AI connections in Glossa, and it takes effect on the next call rather than whenever a token happens to expire.

Connecting Claude

This is the path we have run end to end, on Claude.ai and in Claude Code.

In Claude.ai: open Settings → Connectors, choose Add custom connector, give it a name and the URL https://readglossa.com/mcp, then follow the prompts to Sign in now and Register automatically. Claude sends you to Glossa’s own consent page, you sign in with the Google account you already use for Glossa, and you approve the connection. After that the tools show up grouped as read-only.

In Claude Code, mint a personal access token from Account → AI connections (it is shown once, so copy it then) and add the server:

claude mcp add --transport http glossa https://readglossa.com/mcp \
  --header "Authorization: Bearer glsa_…"

ChatGPT and other clients

Honesty first: we have not completed a ChatGPT connection ourselves. The server implements the same standard ChatGPT’s connectors are built on — OAuth 2.1, dynamic client registration, PKCE, streamable HTTP — so it should work, but “should” is doing real work in that sentence and we will not write step-by-step instructions for a menu we have not walked through. OpenAI’s custom-connector interface also currently requires a paid plan with developer mode enabled, and the menu naming has moved more than once; their own documentation is the reliable source for where the button is today. The URL to paste is the same one: https://readglossa.com/mcp.

The same applies to Gemini, Cursor, open-source agents and anything else with MCP support. Nothing in the server is Claude-specific. If a client implements the protocol, it will find the connector; if it does not connect, we would genuinely like to hear about it.

Things worth asking once it is connected

“Based on what I have actually finished, what should I read next — and is it a step up or a step sideways?” The assistant can see your library and your level, and the recommendations carry their own reasons, so it can argue with them.
“Which of my struggling words share a root or a pattern?” Failed words cluster more than you expect — one verb family, one tense, one false-friend trap. That is a pattern a model is good at spotting and a review queue is not.
“Make me five sentences using only words I have already seen, plus these three I keep failing.” Drilling generated from your own vocabulary rather than a generic frequency list.
“Am I actually reading, or just opening the app?” Pages per day over the last month is an unflattering, useful number, and the assistant will read it back to you without the motivational framing.

A caveat on the last one: reading activity is capped at a 30-day window, because that is how long the underlying per-page log is kept. Ask for a year and you would get a confidently wrong zero for the days whose rows have been swept, so the tool refuses to go that far rather than invent them.

The boring parts

Pro only. The connector is part of the paid tier. The free tier (two books, ten pages) is for trying the reader itself.
Your data goes where you send it. Connecting means the tool results travel to whichever assistant you connected, under that company’s terms. That is the trade the feature is; it is worth making deliberately. See our privacy policy.
Rate limited. 120 calls a minute per account, which an agent in a loop will notice and a human will not.
Sign-in is still Glossa’s. Approval happens on our consent page behind your normal sign-in; the assistant never sees your credentials.

Why this is more than a novelty

The recurring problem with reading in a foreign language is not motivation, it is calibration — picking the next book that is hard enough to teach you something and easy enough to finish. Glossa already does part of that mechanically: glossing the words above your level is what lets you start a book you are not technically ready for. But the judgement call about which book, and what your last two months of reading say about your next two, is a conversation. Now it can be a conversation with something that has the data in front of it.

If you have not read with Glossa yet, start there — upload a book or pick one from the catalog — and connect an assistant once there is a reading history worth asking about.