Agent-ready · WebMCP

WebMCP File Converter

WebConverter is now a WebMCP server: AI agents can convert your files through document.modelContext — and the conversion still runs entirely in your browser.

WebMCP (the Web Model Context Protocol) lets a website hand structured tools directly to AI agents. WebConverter registers its conversion tools so an agent can convert images for you — with no upload, no API key, and no server ever touching your files.

What is WebMCP?

WebMCP is a draft W3C standard from the Web Machine Learning Community Group. It adds document.modelContext to the browser so a page can register tools — each with a name, a description and a JSON Schema — that an AI agent can discover and call. Unlike a classic MCP server there is no separate process and no token: the tool is the page's own JavaScript, so your data never leaves the tab.

Why a File Converter MCP with no upload?

Most AI assistants can only convert a file by uploading it to a third-party API. WebConverter's WebMCP tools run the same WebAssembly pipeline the human interface uses, inside a Web Worker. The agent passes bytes in and gets bytes out — no upload, no rate limit, no cost, near-zero CO₂. It is the most private way for an agent to convert a file.

How an agent uses it

The agent first calls list_supported_formats to learn which formats are valid, then calls convert_image with the file and a target format. The tool returns the converted file as base64 plus a data: URL. Everything is deterministic and described by JSON Schema, so the agent never has to guess.

Privacy and safety by design

The exposed tools are strictly read-only: they take bytes and return bytes. They never write to disk, never make network requests, and never read other tabs. An agent calling these tools has exactly the capability a person clicking “convert” has — and nothing more.

Registered WebMCP tools

These tools are registered with document.modelContext on every WebConverter page and mirrored on a programmatic registry so they work even before native browser support ships.

convert_image

Convert an image (BMP, DDS, GIF, HDR, ICO, JPEG, KTX, PGM, PIC, PNG, PPM, PSD, TGA, WebP, plus HEIC/AVIF via browser decode) to PNG, JPEG, BMP, TGA, HDR, EXR, KTX2 or WebP. Returns base64 + a data: URL. Runs locally, no upload.

images_to_pdf

Combine one or more images into a single PDF, one image per page, in the browser.

images_to_searchable_pdf

Combine images into a PDF and run Tesseract OCR so the resulting PDF has selectable, searchable text on top of the original pixels.

merge_pdfs

Merge multiple PDFs into one (all pages, in order) locally — no upload.

reorder_pdf_pages

Reorder the pages of a PDF and write a new PDF with the given order.

delete_pdf_pages

Drop the listed pages from a PDF, writing a new PDF of the remaining pages in their original order.

extract_pdf_text

Pull text out of a PDF (pdf.js) — plain text by default, or simple Markdown with one heading per page.

convert_document

Convert documents between formats with Pandoc (WASM): DOCX, ODT, RTF, HTML, Markdown, LaTeX, RST, EPUB, MediaWiki, … ↔ Markdown, HTML, plain, LaTeX, RST, AsciiDoc, DOCX, ODT. Lazy ~56 MB download on first use.

convert_audio

Convert an audio file — or extract the audio track from a video — to MP3, OGG, WAV or FLAC. Anything the browser can decode is accepted.

convert_video

Transcode a video to MP4 (H.264 + AAC), WebM (VP9 + Opus) or animated GIF with ffmpeg-wasm. Lazy per-variant download.

trim_video

Cut a clip from startTime → endTime out of a video with ffmpeg-wasm, locally.

remove_image_background

Produce a transparent PNG or WebP version of an image using a tiny U²-Net-P ONNX model plus a deterministic WASM matting pass.

transcribe_audio

Transcribe audio — or the audio track of a video — to text with timestamps using a quantised Whisper model (whisper.cpp WASM). English by default.

list_supported_formats

Enumerate every input/output format and engine across all WebMCP tools so an agent can plan a valid conversion.

Live demo — convert through the WebMCP tool

This calls the exact same convert_image tool an AI agent would call. Pick an image, choose a format, and the conversion runs locally in your browser.

Tip: in Chrome you can enable the native API at chrome://flags/#enable-webmcp-testing. This demo also works without it via the built-in tool registry.

WebMCP File Converter — FAQ

What is the WebMCP File Converter?

It is WebConverter exposed as a WebMCP server. Using document.modelContext, AI agents can call convert_image and list_supported_formats to convert images entirely in the browser — no upload and no API key.

Does it upload my files to a server?

No. The convert_image tool runs the same WebAssembly conversion the UI uses, inside a Web Worker. Your file never leaves the browser tab.

Which browsers support document.modelContext?

It is available behind a flag in recent Chrome builds (chrome://flags/#enable-webmcp-testing) and requires a secure context. WebConverter also exposes a programmatic tool registry so the same tools work before native support ships.

Is the File Converter MCP free?

Yes. There is no backend, no API key and no rate limit — it is as free as opening the page.

What formats can the MCP convert?

Input: BMP, DDS, GIF, HDR, ICO, JPEG, KTX, PGM, PIC, PNG, PPM, PSD, TGA, WebP. Output: PNG, JPEG, BMP, TGA, HDR, EXR, KTX2, WebP.

Is it safe to let an agent call these tools?

Yes. The tools are read-only — bytes in, bytes out. They never write files, make network requests, or read other tabs.

Browse everything WebConverter can convert

Every converter on the site is agent-ready. Explore the full catalogue or read how the WebMCP integration was built.

All conversions