File tools an AI agent can actually use
Most file tools quietly assume a human: an account to create, a CAPTCHA to solve, a file picker to click. An agent gets stuck on the first one.
No sign-up to complete, no CAPTCHA to solve, no upload to wait on. 56 tools for PDF, image, video and audio — each one openable from a URL your agent builds itself.
Six things an agent checks before it can use a tool
No account. There is no sign-up, no login and no session. An agent cannot complete a registration flow, so we removed the need for one.
No CAPTCHA. Nothing here asks you to prove you are human, because on MyTools it makes no difference whether you are.
No upload. Files are processed in the browser. MyTools has no server-side file storage, which removes an entire data-handling review before an agent can act.
A URL you can construct. Append /by-url?url= and a URL-encoded file address. 51 of the 56 tools then open with the file already loaded — no file picker, no drag and drop.
Callable tools. In Chromium with WebMCP, MyTools registers tools on document.modelContext, so an agent calls a function instead of guessing at the DOM.
Free, no quota. Every tool is free and unlimited, and none of them adds a watermark to the file it hands back.
Three ways to drive MyTools
In increasing order of capability. The first works in any browser today.
Open a tool from a URL. Build /image/crop-image/by-url?url= plus a URL-encoded file address, and the page opens ready to run. Repeat ?url= for tools that take several files.
Call the tools directly. Where WebMCP is available, find_mytools_tools, open_mytools_tool and suggest_mytools_pipeline are registered on every page, alongside per-tool run and download tools.
Chain them into a pipeline. Every result carries its type, so the next compatible tool is known: crop, then compress, then combine into a single PDF — without downloading in between.
Machine-readable
Both documents are generated from the same registry that powers the tools themselves, so neither can drift from what the site actually does.
/llms.txt — A summary of the whole site in the llmstxt.org format: what MyTools does, how the URL entry point works, and every tool on one page.
/for-ai-agents/capabilities.json — Every tool with its accepted and produced MIME types, batch support and by-url template. Fetch it once and route correctly from then on.
The tools themselves
56 in total across PDF, image, video and audio. These are the ones agents reach for most.
- Crop Image — Your image has too much content, or a scanned document has wide margins. Remove the unwanted portions in seconds.
- Compress Image — An image is too heavy for the form, the email, or the web page. Compress it without a quality drop you'll notice.
- Merge PDF — You have two or more PDFs and need them as one document. Combine them in the right order in seconds.
- Compress PDF — Your PDF is too heavy to send by email or upload to a form. Reduce its size without losing what matters.
- Rotate Video — Your clip was shot sideways and every player shows it wrong. Fix the orientation in one click — no re-encode, no software to install.
- MP4 to MP3 — You only need the soundtrack of a clip. Extract it as MP3 and skip the video file entirely.
Questions an agent should have answers to
Does using MyTools require an account or an API key?
No. There is no sign-up, no login, no API key and no session. Every tool works on first visit.
Are files uploaded to a server?
No. Processing happens in the browser and MyTools has no server-side file storage. The one exception is the ?url= entry point: to load a remote file, our server fetches the URL you supply so the browser can work on the bytes. It is not stored or retained.
How do I open a tool with a file already loaded?
Append /by-url?url= to the tool's path, followed by a URL-encoded http or https file address. For tools that take several files, repeat the ?url= parameter. 51 of the 56 tools support this; capabilities.json lists which.
Can an agent call MyTools tools directly instead of using the page?
Yes, in browsers that support WebMCP. MyTools registers find_mytools_tools, open_mytools_tool and suggest_mytools_pipeline on document.modelContext on every page, plus run and download tools on individual tool pages. WebMCP is currently a Chromium origin trial, so treat it as available-if-present rather than guaranteed.
Is there a rate limit, a quota, or a cost?
The tools are free and unlimited, and add no watermark. Because the work happens in the browser, throughput is bounded by the machine running it rather than by us.
How do I know which tool handles a given file type?
Fetch /for-ai-agents/capabilities.json. It lists every tool with the MIME types it accepts and produces, whether it handles batches, and its by-url template.
Is the site available in other languages?
Yes, in eight: English, French, German, Spanish, Portuguese, Italian, Indonesian and Arabic. Non-English locales are served under a prefix such as /fr/, and the URL slugs are translated too — /image/crop-image becomes /fr/image/rogner-image.
Can an AI assistant do it for you?
One article per task: what an assistant checks before it acts, the URL it builds, the tools it can call, and where it hands the job back to you.
- Can an AI Assistant Pull the Pages You Need Out of a PDF? — Yes — given a link, it opens the PDF with MyTools, reads the page count, copies out the range you name and saves a new file; when only you know which pages matter, you click them and it does the rest.
- Can ChatGPT Trim a Video for You? — Yes — an AI assistant can open the MyTools video trimmer with your clip loaded, read its real duration, cut it in seconds and save the result. Here is what it checks first, and the one call it hands back to you.
- Can Your AI Assistant Make a PDF Small Enough to Send? — Yes — give it a link and it compresses the PDF in your browser, reports the new size and saves the file; you only step in to judge whether the pages still read well.
- Can an AI Assistant Crop an Image for You? — Yes — with a link it runs the whole job unattended; with a photo on your own machine you drop it in and it takes over. Here is what your assistant checks before it agrees.