Paperzilla
Prepare morning research briefs from focused paper feeds, discover public digests, and read papers with source links through OAuth MCP.
Hosted MCP Server
npx add-mcp 'https://paperzilla.ai/api/mcp'Installs into Claude Code, Codex, Cursor and more
Documentation
A research brief you can talk to.
Every morning, your AI assistant reads what is new in your Paperzilla feeds and tells you which papers are worth your time, and why. Then you ask the questions you would ask a colleague, answered from the papers themselves.
Works with Claude, ChatGPT, and any AI assistant that supports MCP.
Choose a field to see its morning brief
LLM evaluation
Biomedicine
Urban climate
Morning brief · Reliable LLM-as-a-judge evaluation
08:00
Assistant
Good morning. Two new papers in your feed are worth your time.
1
Prompt Lottery in LLM-as-a-Judge
Start here if you rank models with one judge prompt. Rewording the prompt without changing its meaning shifted verdicts, and sometimes reversed model rankings.
2
A fix you can try today. Asking the judge to review its own verdict for bias helped in 20 of 21 model and bias combinations, without extra training.
You
How much worse are LLM judges than people at this?
Assistant
Clearly worse. In a user study with 10 annotators, the judges’ agreement dropped 6 to 9 times more than the humans’ when prompts were reworded. The authors recommend combining verdicts from several prompt variants and trusting the ones that stay consistent.
Example conversation. The papers are real, so you can follow the links and check the answers.
Why a brief
From a list of titles to two good reasons.
Paper alerts tell you what was published. A brief tells you what to read, and why it matters for your work.
Paper alert · 08:00
38 new papers match your keywords
and 29 more
You skim the titles, open a few tabs, and promise yourself you will read them later.
Your brief · 08:00
Two papers worth your time
1
Read first
Why it matters for your work
2
Read next
What to check before you rely on it
→
Your question
Answered from the paper, with a link to the figure
Two papers, a reason for each, and room to ask what you want to know.
Why it works
Paperzilla finds the papers. Your assistant makes them useful.
Each does the part it is good at. Paperzilla watches the sources every day and ranks new papers against your topics. Your assistant goes through that short list with you.
01
It picks, it doesn’t list
Paperzilla narrows thousands of new papers a day to the ones that match your topic. Your assistant narrows those to the few worth reading today.
02
Your assistant knows what you are working on, so “why it matters” is about your project, not a rewritten abstract.
03
It answers follow-ups
Ask about methods, data, or limitations. Answers come from the paper itself, with a pointer to the section or figure.
04
It gets better over time
Your feedback on the papers it brings you goes back to your feed, so the ranking keeps improving.
A week of briefs
Some days it’s two papers. Some days it’s none.
The value builds up over the week. A brief that knows when to stay quiet is one you keep reading.
Mon
Two worth reading
A new method, and a dataset you could reuse.
Tue
A quiet day
Nothing new worth your time. That is the whole brief.
Wed
One that pushes back
A study that questions an assumption in your own work. Better to know now.
Thu
One from next door
A paper from a neighboring field, with a method you could borrow.
Fri
The weekly roundup
One summary across all your feeds, with the week’s must-read on top.
Ask anything
Questions worth asking over coffee.
A brief is the start of a conversation. These are the kinds of questions researchers ask next.
Triage
“I have ten minutes. Which one should I read?”
Methods
“What data did they use, and how much of it?”
Skeptic
“What would a tough reviewer say about this?”
Apply
“Could this work on our data? What would we need?”
Compare
“How does this compare with the other papers in my feed?”
Explain
“Explain the main result as you would to a new PhD student.”
How it connects
Your research feeds, plugged into your assistant.
MCP (Model Context Protocol) is an open standard that lets AI assistants use other apps. Connect Paperzilla once, and your assistant can read your feeds and open the papers in them. No copy-pasting, and no web searches that miss yesterday’s preprints.
Sources
New papers
arXiv, bioRxiv, medRxiv, PubMed, and more
Paperzilla
Your feeds
Ranked against each of your topics, every day
MCP
The connection
Your assistant signs in and reads your feeds
Assistant
Claude, ChatGPT, and others
Picks, explains, and answers your questions
You
08:00
A brief that is ready with your coffee
You sign in with your Paperzilla account and approve what your assistant may do.
Get started
Your first brief in three steps.
01
Create a research feed
Describe your topic in a few sentences. Paperzilla builds the feed and starts collecting matching papers. Already have one? Skip ahead.
02
Connect your assistant
Add the Paperzilla plugin in Claude or ChatGPT, or connect any AI assistant that supports MCP. Then sign in with your Paperzilla account.
03
Schedule your morning brief
Ask for a first brief to see what you get. Then save it as a daily task, and it is waiting for you every morning.
Claude
Plugin and connector
Includes the Morning Brief skill. Schedule it as a recurring task.
ChatGPT
Plugin
Add the Paperzilla plugin and sign in with your Paperzilla account.
Any MCP assistant
MCP connection
OpenClaw, or any other assistant that supports MCP. Sign in with your Paperzilla account and schedule briefs in your own setup.
Google Gemini
Planned
Support is on the way.
FAQ
Common questions.
Why not just ask my assistant what is new, without Paperzilla?
You can, but web search is a poor way to keep up with a field. It favors popular and older pages, can easily miss yesterday’s preprints, and turns up different results each time. Paperzilla checks arXiv, bioRxiv, medRxiv, PubMed, and more every day, ranks each new paper against your topic, and keeps track of when each one arrived. Your assistant starts from that complete, ranked list instead of whatever a search happens to find.
What do I need?
A Paperzilla account with at least one research feed, and an AI assistant that supports MCP connectors. Scheduled tasks may need a paid plan with your assistant.
What can my assistant see and do?
It reads your feeds and the papers in them. Anything that changes your account asks for your approval first, and you can disconnect it at any time.
Does it read the whole paper?
When you ask a detailed question, your assistant can open the full text where it is available, and point you to the section or figure it used.
What happens on a day without new papers?
The brief says so. A quiet day is useful information, and it gives you your morning back.
Can I follow several topics?
Yes. Create a feed for each topic, then get one brief across all of them or one per feed.
Start with one topic. Read your first brief tomorrow morning.
Create a feed for the question you care about, connect your assistant, and let the new papers come to you.
Building your own workflow? See Paperzilla for agents →