Get the review comments out of an Overleaf paper, into Markdown and JSON you can read, keep, commit, and hand to an AI assistant.
Overleaf leaves comments out of both the source download and the Git sync, so the only way to work through them is inside the editor, one at a time. On a paper with ninety comments from three co-authors that stops being workable.
It also writes a PDF of your paper with the comments highlighted on the words they were written about, coloured by who wrote them. No LaTeX to install and nothing to compile. See Comments inside the PDF.
## Summary
- **Threads:** 2 (2 open, 0 resolved)
- **Tracked changes:** 0
- **Most active reviewers:** A. Reviewer (2), Co Author (1)
### § Method
**Line 142** — 2 comments
> …We propose **▸a novel framework◂** for measuring perceived quality across conditions.
**C014** _open · 1 reply_ — "a novel framework"
- **A. Reviewer** · 2026-02-25 06:13 UTC: Needs a citation. Which studies show this?
- ↳ **Co Author** · 2026-02-26 10:00 UTC: Agree, Smith 2023 would work here.
**C015** _open_ — "coherent"
- **A. Reviewer** · 2026-02-27 13:46 UTC: This word is doing a lot of work. Define it.Every comment sits in its section, next to the words it was attached to, with its replies underneath and a stable id you can point at.
If you do not write code, download the app for your computer from the latest release, unzip it, and open it. There is nothing else to install, and no Python needed.
| Your computer | Download |
|---|---|
| Mac | OverleafCommentsExport-macOS.zip |
| Windows | OverleafCommentsExport-Windows.zip |
| Linux | OverleafCommentsExport-Linux.zip, then chmod +x "Overleaf Comments Export" |
The first time you open it, macOS or Windows will warn you about an app it has not checked. That is because signing an app costs money every year, not because anything is wrong. On macOS, right-click the app and choose Open. On Windows, click More info, then Run anyway.
If you use Python, it is one command:
pip install overleaf-comments-export
overleaf-comments-export --guiThat opens the same window. Paste your project link, choose a folder, press the button.
The window is not a reduced version of the tool. Everything below is in it, including the reply letter, the comments-in-the-PDF option, and the tidy output for version control. If you prefer the terminal:
overleaf-comments-export \
--project-url https://www.overleaf.com/project/YOUR_PROJECT_ID \
--out ./paper-commentsPython 3.10 or newer, tested up to 3.14, on macOS, Windows, and Linux.
| File | What it is |
|---|---|
comments-<date>.md |
The readable export, or comments.md with --stable. Grouped by file, then section, then line. Stable ids C001, T001. |
comments.json |
The same data, structured. Threads, anchors, tracked changes, source context. |
comments.jsonl |
One self-contained record per comment, for pipelines. |
agents.md |
A short brief telling an AI assistant how to read the other two. |
response-letter.md |
Optional. A point-by-point reply document with a blank slot under every open comment. |
commented.pdf |
Optional. Your paper as Overleaf builds it, with the comments highlighted on the text. |
annotated/ |
Optional. Your LaTeX with the comments embedded, if you would rather compile it yourself. |
by-reviewer/<name>.md |
Optional. One file per reviewer, so you can work through them one person at a time. |
This is what the export is shaped for. Hand an assistant the folder and it can work on any comment by id, because it already knows the quoted passage, the section it lives in, and what the replies said.
Some things that work well.
- Draft a reply to C014 that answers the reviewer and says what we changed.
- Which comments are asking for the same thing? Group them.
- Fill in the response letter for every comment from A. Reviewer.
Tracked changes come through too, as insertions and deletions with the
surrounding text, so a deletion reads as before ~~removed~~ after.
--pdf # a PDF of your paper with the comments in it
--response-letter # draft a point-by-point reply document
--annotated-tex # a copy of your source with the comments embedded
--annotate-style todonotes # put those comments in the margin instead
--per-reviewer # one report per reviewer
--reviewer "Emma" # only threads this person touched
--no-resolved # only what is still open
--no-open # only what is already resolved
--no-changes # skip tracked changes
--render-mode detailed # more source context around each comment
--stable # output that only changes when the comments change
--cookie "PASTE" # sign in with a pasted cookie
--base-url https://... # a self-hosted Overleaf
--no-jsonl # skip comments.jsonl
--include-raw # keep the untouched Overleaf data in comments.json
--version # which version this isRun overleaf-comments-export --help for the full list.
The window follows whether your computer is set to light or dark. There is an Appearance picker in the top right if you would rather fix it one way.
pip install 'overleaf-comments-export[pdf]'
overleaf-comments-export --project-url <link> --out ./review --pdfThis writes commented.pdf: your paper exactly as Overleaf builds it, with
each comment highlighted on the words it was written about, in the colour of
whoever wrote it. Hover a highlight and the comment appears. This is what
people have been asking Overleaf for since 2023.
There is nothing to install beyond that one line, and nothing to compile. The PDF Overleaf already built is fetched and the highlights are written into it, so it has your real document class, your real figures, and your real bibliography, and it cannot fail to build. Your project is never modified.
If Overleaf has no build to hand, the tool asks it to compile, so this works whether or not you have opened the project recently.
Use a browser. Chrome, Edge, Firefox and Safari all show the highlights and the comment on hover, and none of them need anything installed. Adobe Acrobat Reader works too, and shows the comments in a side panel.
Preview on a Mac shows the highlights but not the comments. Preview does not display the note attached to a highlight, so hovering does nothing. This is Preview, not the file, and it affects any commented PDF, not just these. Drag the file onto a browser window instead.
Whatever you open it in, the last pages list every comment in full with the page it is on, so nothing depends on your reader supporting annotations.
A PDF does not know where the LaTeX went, so the commented words are found by
matching the source against the text on the page. Citations and cross
references are dropped, because they print as numbers. Maths is treated as
unknowable and the words either side of it are used instead. Line-break hyphens
are ignored, so a comment on force-feedback still matches when it was set as
force- and feedback on two lines. A phrase that appears more than once is
resolved by reading in step with the source order.
On the paper this was built against, 82 of 83 comments landed on the right
words. The one that did not was written on a \abstract command, which prints
nothing. A comment that cannot be placed is listed rather than put somewhere
plausible.
Papers split across several .tex files are handled; every file that carries a
comment is matched against the same pages.
--annotated-tex writes a copy of your LaTeX into annotated/ with the same
highlighting, to compile on Overleaf or on your own machine.
overleaf-comments-export --project-url <link> --out ./review --annotated-texWe propose \pdfmarkupcomment[markup=Highlight,color=ocehlA,author={A. Reviewer}]%
{a novel framework}{[C014] A. Reviewer: Needs a citation. | Reply: Smith 2023 would work}
for measuring quality.A key at the top of the document says whose colour is whose. Where more than one comment covers the same words, that stretch gets its own colour and a single note listing all of them. Where the same person has commented twice on overlapping words, their own colour darkens rather than changing, so it does not read as somebody else. Resolved comments are struck through in grey.
Some spans cannot be highlighted, because the package that draws the markup rebuilds the text and fails on maths, on LaTeX commands, and across paragraph breaks. Those become a pin at the nearest safe point instead. Every comment, highlighted or not, is listed on a page at the end, which is also what you get when the PDF is printed, since readers do not print annotations.
--annotate-style pdfcomment goes back to a pin at each spot, and
--annotate-style todonotes puts notes in the margin.
Your own source is never touched. The annotated copies keep their original
names and folder structure, so \input still resolves if you upload the folder.
Reviewer text is made safe before it goes anywhere near the file. A comment
containing 50% of $x$, a stray backslash, a Greek letter, or punctuation typed
on a Chinese keyboard cannot break the build, because everything written out is
reduced to plain ASCII first. Overleaf compiles with pdflatex, which stops the
whole document on a single character it cannot encode.
--response-letter writes the document you actually have to hand in. Every
open comment, grouped by whoever raised it, with the passage it refers to,
where it sits, the discussion so far, and blank lines for your answer.
### C102 — § Method > Measures (line 192)
**Referring to:** "drawing selected quotes to point at possible sources of mismatch"
**Comment:**
> more than just mismatch i would say
**Response:**
_TODO_
**Change made:**
_TODO — what changed, and where._The ids match comments.json, so you can ask an assistant to draft any point by
id and it already knows the quote and the location.
By default each run writes comments-<date>.md, so you end up with one file per
run. That is fine for a one-off export and useless for version control.
--stable writes a single comments.md with no timestamps in it. Re-run it and
the file does not move unless the comments actually moved.
overleaf-comments-export --project-url <link> --out ./review --stable
git add review/comments.md
git commit -m "reviewer comments as of today"Next time you pull the comments, git diff shows the new ones and nothing else.
thread_count: 41
+thread_count: 42
+**C042** _open_ — "the axis labels"
+- **A. Reviewer** · 2026-08-14 09:12 UTC: units are missing hereIf your university runs its own Overleaf rather than using overleaf.com, tick
"My university runs its own Overleaf" in the window and put in its address, or
pass --base-url https://overleaf.my-university.edu on the command line.
Comments work. Tracked changes do not, because Overleaf makes them part of
Server Pro rather than the free Community Edition, so there are none to export.
Self-hosted Overleaf names its session cookie overleaf.sid rather than
overleaf_session2, which the tool handles either way.
This path is less tested than overleaf.com, so if something does not work, please say so.
The tool needs the session your browser already has. It reads that session and nothing else, and it only ever reads.
| How | Notes |
|---|---|
| Safari or Firefox | Reads the cookie file directly. No password prompt. |
| Paste it yourself | Works on every computer and browser. The window shows you how, step by step. |
| Chrome, Edge, Brave | Works on macOS but asks for your Keychain password every run. Chrome 127 and newer on Windows cannot be read at all, so use the paste option there. |
To paste it, open Overleaf, press F12, go to Application or Storage, then
Cookies, then https://www.overleaf.com, and copy the value of
overleaf_session2. Treat it like a password. It stops working when you sign
out. Nothing is stored unless you tick the box that says so.
overleaf-comments-export --project-url <link> --out ./out --cookie "PASTE_HERE"| What you see | What it means |
|---|---|
| Could not look up www.overleaf.com | This computer is offline, or a VPN is in the way. Nothing to do with your project. |
| Overleaf refused the request | The session expired. Sign in again in your browser and re-run. |
| Could not read Overleaf cookies | Use the paste method above. |
| The window cannot open | Python's Tk toolkit is missing. The message tells you what to install for your system. |
Anything else, please open an issue. Never paste your session cookie into an issue.
Several people hit this same wall independently. Different approaches suit different people, so here are the others.
- adakite/extract-overleaf-comments works from a saved copy of the page, and can put comments in the PDF margins.
- salokr/overleaf-comment-exporter is a browser extension, so there is nothing to install on the command line.
- IcarusLemon/overleaf-comments-cli is another command line exporter, driving a real browser under the hood.
All of them exist because of this three year old request. An official export would make every one of us unnecessary, which would be the better outcome.
This is an unofficial tool. It uses Overleaf's internal endpoints, which are undocumented and can change with any release. It identifies itself honestly in every request, backs off when asked to, and cannot modify your project. Even so, it may stop working one day without warning. If that happens, please say so in an issue, because you may be the first to notice.
It is not affiliated with or endorsed by Overleaf. Use it on projects you have access to, in line with Overleaf's terms.
Feedback shapes what gets built next.
- Something broke, or the output was wrong. Open an issue. You do not need to be a programmer. Paste what the tool said and that is plenty.
- Want it to do something it does not. Suggest a feature. Describe the task, not only the feature. The real task usually leads somewhere better.
- A question, or something you built with it. Discussions.
- Code. See CONTRIBUTING.md. Two minutes to get the tests running, and there are items marked help wanted in ROADMAP.md.
What is coming next is in ROADMAP.md. What already changed is in CHANGELOG.md.
Maintained by Shivang Gupta, who wrote it to handle the review comments on his own papers.
MIT. See LICENSE.