pdfmcp vs IHatePDF: the on-device PDF tools compared
IHatePDF (ihatepdf.cv) pioneered 100% client-side PDF tools — no server, no uploads, all free. pdfmcp shares that model and adds the layer a website alone can't offer: the same tools as a local MCP server, a local REST API and a library for AI agents and scripts. An honest comparison.
The quick verdict
This is the closest comparison on this site, because the web tools share a philosophy: IHatePDF runs everything in your browser via WebAssembly — there is no processing server, nothing is uploaded, it works offline as a PWA, and it is entirely free with no sign-up and no watermark. pdfmcp's web tools work the same way, for the same reasons.
The difference is what sits behind the website. IHatePDF is web-only, so there is nothing for an AI agent or a script to call. pdfmcp adds exactly that layer — on your machine, not ours: a local MCP server, a local REST API (pdfmcp serve, on localhost) and a Python library running the same operations, so agents and scripts get the tools without any file ever leaving your device.
| pdfmcp | IHatePDF | |
|---|---|---|
| Where files are processed | On your device — browser or local package | On your device, in your browser |
| MCP server for AI agents | Yes — free, runs locally on your machine | None — web-only, nothing for an agent to call |
| API / library for scripts | Yes — local REST API (localhost) and local Python library, no API key needed | None |
| Sign-up | Never — there are no accounts | Not needed |
| Price | Free forever — no paid tier | Free — no paid tier |
| Watermark on output | Never | Never |
| Tools | 74 operations | 46 tools listed in its open-source repository (site claims more) |
| Works offline | Yes — web tools once loaded; local package entirely | Yes (PWA) |
| Heavy jobs (large OCR, Office conversion) | Yes — local package on your machine (engines fetched/installed once) | No — bounded by your browser tab |
| AI chat with PDF | No — deliberate non-goal | Yes (extracted text is sent to Gemini) |
| Unique extras | AI-provenance metadata stripping; text as JSON with bounding boxes | PDF→audio, audio→PDF, invoice/billing generators, P2P file share, privacy scanner |
| File retention | Nothing is uploaded, so there is nothing to delete | Nothing is uploaded, so there is nothing to delete |
Privacy, side by side
On web tools, this is a tie — and we are happy to say so. Both process files on your device; neither uploads your documents; there is nothing on a server to delete. IHatePDF's one exception is its Chat with PDF feature, which sends extracted text to Google's Gemini; pdfmcp offers no AI chat, precisely because it would require sending content to an AI provider.
pdfmcp's local package follows the same rule off-browser: it processes files on your machine, has no telemetry, and never phones home — its only downloads are one-time, checksum-verified engine fetches (qpdf, OCR language data) from official upstream releases, which you can disable. If you want zero server involvement of any kind, both sites' web tools give you exactly that.
Pricing and limits
Another tie: both are entirely free, with no paid tier, no sign-up and no watermark. The practical limits are the same in kind: a purely on-device tool is bounded by the device in your hand — memory and CPU — which is why heavy OCR or a 400-page conversion can strain a phone.
pdfmcp's local package moves that bound from the browser tab to your whole machine: the same operations run as a desktop-class process, with heavy engines fetched or installed once (qpdf downloads verified on first use; Ghostscript and Tesseract are system installs; LibreOffice likewise for Office conversion). Same price — $0, forever — with a bigger local engine available when the browser tab runs out of road.
Who should pick which
Pick IHatePDF if: you want its unique extras — PDF to audio, audio to PDF (on-device transcription), GST invoice and POS billing generators, a P2P file-sharing tool, or its privacy-scanner report — or you simply prefer a single-purpose site with no installable anything, run as an open-source project you can inspect on GitHub.
Pick pdfmcp if: your AI agent needs to call PDF tools (a local MCP server is the whole point); you want the same operations over a local REST API on your own machine or as a Python library for scripts; a job is too big or slow for a browser tab and should run as a full local process; or you want agent-first outputs like text as JSON with bounding boxes and AI-provenance metadata stripping alongside the everyday tools.
Frequently asked questions
Is pdfmcp just a clone of IHatePDF?
The web tools share IHatePDF's insight — run everything on-device — and we credit it. pdfmcp's reason to exist is the layer a website alone can't have: the same tools as a local MCP server, a local REST API and a Python library for agents and scripts, still with no upload anywhere.
Which is more private?
For web tools, identical: files never leave your device on either site. pdfmcp's local package keeps the same rule off-browser — processing on your machine, no telemetry, only one-time verified engine downloads.
Can my AI agent use IHatePDF?
No — it has no API or MCP endpoint for an agent to call. That gap is precisely what pdfmcp fills: the same operations, exposed as MCP tools with typed schemas, running locally on your machine.
Which has more tools?
pdfmcp lists 74 operations. IHatePDF's open-source repository enumerates 46 tools (its site claims 60+), including extras pdfmcp doesn't offer, like PDF-to-audio and invoice generators.