WEBMCP · AI-AGENT READY
Your AI agent can compress images here
ImgFlow ships native WebMCP support. Open the tool page in a WebMCP-capable browser — Chrome 149+ (origin trial) or the ChatGPT desktop browser — and the page automatically registers six tools that let the built-in AI agent add images from URLs, configure compression (format, quality, target KB), monitor progress and download finished files. The agent drives the exact same on-device pipeline as your clicks do: this site has no upload endpoint for it to send images anywhere.
Start processingWhat WebMCP is
WebMCP is a Chrome proposal (2026 draft) that lets a web page expose its capabilities as tools to the AI agent built into the browser. The page calls document.modelContext.registerTool(); the agent can then operate the page on the user's behalf, chaining tools to complete a task. It is a browser API, not a network protocol — tools run inside the page, in your browser, and the human stays in the loop: the browser asks for confirmation before consequential actions.
The six tools ImgFlow registers
Tool names and parameters are the exact identifiers the agent sees (the spec requires writing them in English, so they are identical in every language).
| Tool | Input | What it does |
|---|---|---|
add_image_from_url | url (required), name | Adds an image from an https:// or data: URL to the queue; compression starts automatically. |
set_compression_settings | format, targetFormats, mode, quality, targetKB, maxEdge, keepIcc, pngQuantize, start | Configures output format(s), quality (1-100) or target size in KB, max pixel dimension, ICC and PNG quantization; start:true reprocesses the whole queue. |
get_compression_status | — (read-only) | Returns running flag, counts by status, bytes saved, the first 8 results and current settings. |
download_image | name (required) | Downloads one finished image (exact or partial name match) to the default download folder. |
download_all_results | — | Downloads all finished images as a single ZIP. |
reset_queue | — | Clears the queue and all results; marked consequential, so the browser asks for your confirmation first. |
How to use it
- 01Open the tool page
Visit the ImgFlow compressor in a WebMCP-capable browser: Chrome 149+ with the origin trial enabled (or the chrome://flags/#enable-webmcp-testing flag), or the ChatGPT desktop browser.
- 02Ask in natural language
Tell the agent what you need, e.g. "compress these three URLs to WebP under 200 KB and give me a ZIP". The agent plans and calls the tools itself.
- 03Confirm and collect
Results land in the browser's default download folder. Destructive actions (clearing the queue) always require your confirmation.
Example prompts
Prompts that work well on this site:
Compress https://example.com/photo.jpg to under 200 KB and download it.Convert these images to AVIF at quality 75, longest edge 2560, then download all as a ZIP.What are the current compression settings and queue status?
Privacy: the agent changes who clicks, not where images go
WebMCP tools execute inside the page and call the same local WebAssembly engine as the buttons do. This site has no upload endpoint at all, so there is nowhere for an agent (or anyone) to send your images. Remote URLs are fetched by the page itself for processing, and output files strip EXIF/GPS metadata exactly as before.
FAQ
Which browsers support WebMCP today?
WebMCP ships as an origin trial in Chrome 149+ (enable chrome://flags/#enable-webmcp-testing for local testing) and is supported in the ChatGPT desktop browser. In browsers without WebMCP the site works normally — the tools simply are not registered (progressive enhancement, zero extra bytes loaded).
Can the AI agent upload my images somewhere?
No. The tools run the same on-device pipeline as manual clicks, and this site has no image upload endpoint at all. Verify it in DevTools: no image request leaves during processing, and it keeps working offline.
Do I need to install an MCP server or a browser extension?
No. WebMCP is a native browser API — the page registers tools with document.modelContext directly. Nothing to install, nothing to configure, no separate server involved.
Which agent actions require my confirmation?
reset_queue is registered with the consequential hint, so the browser requires explicit user confirmation before it runs. Everything else (adding images, changing settings, downloading results) is as low-risk as clicking the same buttons yourself.
Does it cost anything or need a sign-up?
No. Like everything on ImgFlow, the agent path is free, unlimited and account-less.