Inspiration
“...savour in a momentary glimmer of consciousness the sleep into which were plunged the furniture, the room, that whole of which I was only a small part and whose insensibility I would soon return to share.”
I stared at this sentence for three minutes.
I understood every single word. I still did not understand the sentence.
Damn it, Proust.
While reading In Search of Lost Time, I realized that vocabulary was not my problem. Syntax was. My workaround was to leave the reading app, paste the sentence into ChatGPT, and read a long explanation. After doing this a few times, the interruption became exhausting.
And I did not want a summary.
I wanted Proust.
Reading literature is not just about extracting meaning. It is also about rhythm, structure, metaphor, punctuation, and the way a sentence unfolds.
I wanted just enough help to see how a difficult sentence works, without replacing it with AI-generated summaries.
That is why I built Proust’s Reader.
What it does
Proust’s Reader is an EPUB reader. It helps you understand a difficult sentence by untangling its structure inside the original text.
Highlight a confusing sentence, ask your agent for help, and it can use WebMCP tools to:
Mark the spine — reveal the main subject, verb, and core structure.
Unfold nested structure — show buried clauses and phrases, what they modify, and what they mean in context.
Reorder the syntax — temporarily show the sentence in a more familiar word order.
Show the plain meaning — only if you are still completely lost.
The idea is to give progressively stronger hints instead of immediately flattening the sentence into a paraphrase.
The original sentence always stays at the center.
No separate AI explanation to mentally map back onto the book. No summaries. No turning Proust into bullet points.
Just enough scaffolding to understand the sentence. And get back to reading.
How I implemented WebMCP
I exposed four WebMCP tools from the reader: mark_spine, unfold_nested_structure, reorder_syntax, and show_plain_meaning.
When the reader highlights a sentence, the agent can discover these tools and invoke the one that matches the user’s request. The tool receives the selected sentence and nearby context, and the result is rendered back into the reader so the original sentence can be annotated, unfolded, reordered, or paraphrased directly on the page.
I intentionally exposed several focused tools rather than one generic “explain this sentence” action, so the agent can choose how much assistance to provide.
Why WebMCP fits in Prost's Reader
For my case, WebMCP is especially useful because it helps preserve an immersive reading experience, even when I run into a sentence with a monstrous grammatical structure. (Damn it, Proust!)
Instead of copying the sentence, leaving the book, and asking an AI to explain it somewhere else, I can ask for help and let the agent operate directly on the sentence I am already reading.
More importantly, WebMCP makes the explanation interactive and visual.
The agent can mark the grammatical spine, unfold nested phrases and clauses, or temporarily reorder the syntax. All happen inside the original text.
That makes it much easier to see how the sentence is constructed, instead of reading another explanation and trying to map it back onto the book in my head.
WebMCP makes this simple goal possible: get unstuck without leaving the reading experience.
How people and agent collaborate in Prost's Reader
People still do the reading. They decide when a sentence feels difficult, what kind of help they want, and when they are ready to return to the original text.
The agent provides the scaffolding.
It can use the reader’s WebMCP tools to reveal the sentence spine, unfold nested structures, reorder difficult syntax, or show a plain-English meaning when needed.
Together, the reader and agent can work through a difficult sentence step by step without turning the experience into a separate AI conversation.
The reader keeps control of the interpretation. The agent simply makes the structure easier to see.
Instead of replacing difficult literature, they can now work through it together inside the text.
Built With
- chatgpt
- codex
Log in or sign up for Devpost to join the conversation.