You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Proposal: Add a community showcase or related projects section #2123
It provides a simple UI for uploading supported files, converting them to Markdown, previewing the result, copying the Markdown, and downloading it as a .md file.
I was wondering whether the MarkItDown repository would be open to adding a small "Community Projects", "Related Projects", or "Built with MarkItDown" section in the README. This could help users discover practical tools built on top of MarkItDown, and also make it easier for community projects to link back to the official repository.
If this sounds acceptable, I would be happy to submit a PR adding such a section and include my project as one example.
+1 to this. I also built a project on top of MarkItDown and would benefit from a "Built with MarkItDown" section.
It's is free, MIT-licensed, local-first desktop app for MarkItDown. The conversion engine is MarkItDown under the hood; what it adds is OCR for scanned PDFs, batch processing, cleanup modes, and a reliability/diagnostics layer.
Repo: https://github.com/ibrahimqureshae/mdflux
I also posted a short write-up in the Show and tell discussion: #2154
Happy to submit a PR for the section if the team is open to it. Could list projects.
+1, and I'd like to offer a second concrete example plus some actual help getting it over the line.
I built a Claude Code hook on top of MarkItDown (claude-markitdown-hook, MIT) that converts referenced documents before the model sees them, so an AI assistant reads markdown instead of rendering every PDF page as an image.
The reason I think a section like this earns its place, beyond discovery: the integrations built on MarkItDown keep re-learning the same lesson independently. In my case it was that markitdown correctly exits 0 when a PDF has no text layer — finding nothing genuinely isn't an error — and every downstream integration that reads that exit code as "we got the text" hands its user an empty file. Mine did, until I fixed it. A visible list of integrations is also a place where those lessons become findable instead of being rediscovered one project at a time.
If it's useful, I'm happy to co-write the PR — including a short "things to know when building on MarkItDown" line under the section, so it's contributing something to readers rather than only listing links. Entirely up to the maintainers whether that scope is welcome; happy to submit the plain list version instead.
If a related-projects section is added, FolioQueue would fit. Most MarkItDown wrappers focus on one-shot batch conversion (some add OCR or a GUI); FolioQueue instead focuses on maintaining a Markdown mirror over time — SHA-256-based incremental skip so reruns only convert changed files, per-file checkpoints for interruption-safe resume, and protection for hand-edited output that even --force won't overwrite. MIT, installable via pip install folioqueue. I'm its maintainer.
+1 for a small Community Projects / Related Projects section.
I maintain [MarkItDown 2Ways](https://github.com/Nolane-x/markitdown-2ways), a community fork of Microsoft MarkItDown that keeps the original one-way conversion workflow available while extending it toward bidirectional, preservation-first document editing.
The current v1.0.0 release focuses on a few things that are intentionally different from a normal converter wrapper:
an evidence-driven DocumentIR and typed edit model;
source-preserving round trips for supported native formats;
explicit native ownership and capability checks before mutation;
fail-closed behavior when safe writeback cannot be proven;
17 native format families;
12 additional derived semantic surfaces that remain explicitly non-writeback;
adversarial capability tests, compatibility goldens, and deterministic engineering-budget gates.
A key design rule is that successfully parsing a document does not automatically grant permission to write it back. Before producing edited native bytes, the system re-checks source identity, native ownership, supported operations, and preservation constraints.
The fork is also kept synchronized with upstream MarkItDown; it currently includes upstream changes through commit 945314a45ddbe02935f2fd287b797dc0ba4a01e4.
This is a community-maintained fork, not an official Microsoft project and not endorsed by the MarkItDown maintainers.
If a Related Projects / Built with MarkItDown section is accepted, I think MarkItDown 2Ways would fit well as an experimental bidirectional and preservation-oriented extension of the MarkItDown ecosystem.
MdPipe (https://github.com/gdols/mdpipe, MIT) is a Windows desktop app and CLI on top of MarkItDown. It's a single .exe with nothing to install: it sets up its own Python on first run, and then you drag files or folders onto it and get the .md next to each one. Batch conversion, English and Spanish UI, and it updates itself when there's a new release.
+1 for a related-projects section. One more candidate, from the other direction: docling.rs (https://github.com/docling-project/docling.rs, MIT) is a Rust document converter whose Python package implements MarkItDown’s API (from docling_rs.markitdown import MarkItDown), so code written against MarkItDown runs on it unchanged. It is not built on MarkItDown and not affiliated with it; the overlap is the API and the set of formats. I maintain it. Happy to be listed under “compatible alternatives” rather than “built with”, whichever fits the section.
Activity
ibrahimqureshae commentedon Jun 22, 2026
+1 to this. I also built a project on top of MarkItDown and would benefit from a "Built with MarkItDown" section.
It's is free, MIT-licensed, local-first desktop app for MarkItDown. The conversion engine is MarkItDown under the hood; what it adds is OCR for scanned PDFs, batch processing, cleanup modes, and a reliability/diagnostics layer.
Repo: https://github.com/ibrahimqureshae/mdflux
I also posted a short write-up in the Show and tell discussion: #2154
Happy to submit a PR for the section if the team is open to it. Could list projects.
Built with MarkItDown
Thanks for considering it.
AndrewAvery7 commentedon Aug 7, 2026
+1, and I'd like to offer a second concrete example plus some actual help getting it over the line.
I built a Claude Code hook on top of MarkItDown (claude-markitdown-hook, MIT) that converts referenced documents before the model sees them, so an AI assistant reads markdown instead of rendering every PDF page as an image.
The reason I think a section like this earns its place, beyond discovery: the integrations built on MarkItDown keep re-learning the same lesson independently. In my case it was that
markitdowncorrectly exits 0 when a PDF has no text layer — finding nothing genuinely isn't an error — and every downstream integration that reads that exit code as "we got the text" hands its user an empty file. Mine did, until I fixed it. A visible list of integrations is also a place where those lessons become findable instead of being rediscovered one project at a time.If it's useful, I'm happy to co-write the PR — including a short "things to know when building on MarkItDown" line under the section, so it's contributing something to readers rather than only listing links. Entirely up to the maintainers whether that scope is welcome; happy to submit the plain list version instead.
GokouRuri43 commentedon Sep 10, 2026
If a related-projects section is added, FolioQueue would fit. Most MarkItDown wrappers focus on one-shot batch conversion (some add OCR or a GUI); FolioQueue instead focuses on maintaining a Markdown mirror over time — SHA-256-based incremental skip so reruns only convert changed files, per-file checkpoints for interruption-safe resume, and protection for hand-edited output that even
--forcewon't overwrite. MIT, installable viapip install folioqueue. I'm its maintainer.Nolane-x commentedon Sep 19, 2026
+1 for a small Community Projects / Related Projects section.
I maintain [MarkItDown 2Ways](https://github.com/Nolane-x/markitdown-2ways), a community fork of Microsoft MarkItDown that keeps the original one-way conversion workflow available while extending it toward bidirectional, preservation-first document editing.
The current v1.0.0 release focuses on a few things that are intentionally different from a normal converter wrapper:
DocumentIRand typed edit model;A key design rule is that successfully parsing a document does not automatically grant permission to write it back. Before producing edited native bytes, the system re-checks source identity, native ownership, supported operations, and preservation constraints.
The fork is also kept synchronized with upstream MarkItDown; it currently includes upstream changes through commit
945314a45ddbe02935f2fd287b797dc0ba4a01e4.Repository: https://github.com/Nolane-x/markitdown-2ways
v1.0.0 release: https://github.com/Nolane-x/markitdown-2ways/releases/tag/2ways-v1.0.0
This is a community-maintained fork, not an official Microsoft project and not endorsed by the MarkItDown maintainers.
If a Related Projects / Built with MarkItDown section is accepted, I think MarkItDown 2Ways would fit well as an experimental bidirectional and preservation-oriented extension of the MarkItDown ecosystem.
gdols commentedon Sep 21, 2026
+1 for a section like this.
MdPipe (https://github.com/gdols/mdpipe, MIT) is a Windows desktop app and CLI on top of MarkItDown. It's a single .exe with nothing to install: it sets up its own Python on first run, and then you drag files or folders onto it and get the .md next to each one. Batch conversion, English and Spanish UI, and it updates itself when there's a new release.
artiz commentedon Oct 16, 2026
+1 for a related-projects section. One more candidate, from the other direction: docling.rs (https://github.com/docling-project/docling.rs, MIT) is a Rust document converter whose Python package implements MarkItDown’s API (from docling_rs.markitdown import MarkItDown), so code written against MarkItDown runs on it unchanged. It is not built on MarkItDown and not affiliated with it; the overlap is the API and the set of formats. I maintain it. Happy to be listed under “compatible alternatives” rather than “built with”, whichever fits the section.