For immediate release · August 31, 2026
Smol now speaks MCP — so your AI agent can do the file work.
Claude Code and Codex can now call Smol directly to compress, convert, upscale, resize, and strip metadata from images, video, PDFs, and audio — with precise control over every setting, and every file processed locally on your Mac.
Smol now supports MCP — the Model Context Protocol — letting AI coding tools use Smol’s local media-processing capabilities directly. The update turns a drag-and-drop Mac utility into a callable tool layer for AI agents, without moving a single file off the user’s machine.
AI coding agents have become good at writing and refactoring code, but the assets that ship alongside that code — hero images, screenshots, demo videos, PDF exports, design handoffs — have remained a manual step. An agent could describe what needed to happen to a 4 MB PNG; it could not reliably make it happen. What existed instead was a generic “compress” action with opaque defaults, or a shell command against whatever encoder happened to be installed, producing results that drifted between runs.
Smol’s MCP layer replaces that with something precise. Agents can perform structured image, video, PDF, audio, metadata, conversion, and upscaling tasks — choosing formats, quality targets, compression mode, output location, metadata removal, resize and upscale settings, and batch behaviour. They can plan, run, monitor, and cancel jobs, and read back structured results describing exactly what happened.
“Agents got good at writing the code. The assets around it stayed manual. We gave the agent a real finishing layer — one that runs on your Mac, keeps your files there, and does the job the same way every time.”
Setup is guided from a new in-app area called “Smol for AI”, which shows a live connection status for each client and offers a one-click “Copy full AI guide” so an agent can be handed the complete tool surface in a single paste. Claude Code and Codex are supported out of the box.
The tool surface
Four tool families, not one generic action.
Everything the app can do is exposed to the agent, named the same way it is named in the interface.
Fast, exact compression for images, PDFs, video, and audio.
The agent sets an explicit quality target and compression mode rather than asking for a vague reduction. Production-grade encoders — the same class of tools that run behind cloud services — execute the job on the local machine.
Detail-first image compression guided by visual analysis.
Smol maps faces, text, edges, and high-detail regions before encoding, then protects those areas while compressing flat backgrounds harder. Agents use it when the asset has to stay visually intact at a much smaller size.
Local conversion across images, documents, video, audio, tables, and fonts.
HEIC to JPEG, PNG to AVIF, MOV to MP4, FLAC to MP3, DOCX to PDF, XLSX to CSV, and dozens more — without a browser upload or a per-file quota.
AI upscale 2× or 4×, strip metadata, resize, and automate batches.
Enlarge assets with a local machine-learning model, remove EXIF, GPS, IPTC, and XMP data, and apply one instruction across a folder of hundreds of files.
Granular control
The agent chooses the settings, not a preset.
- Format and codec
- JPEG, PNG, WebP, AVIF, HEIC, RAW, PDF, MP4, MP3 and more.
- Quality target
- An explicit target rather than an opaque default.
- Compression mode
- Classic for speed and precision, Neural for detail retention.
- Output location
- Where results land, and how they are named.
- Metadata removal
- Strip EXIF, GPS, IPTC and XMP — with or without recompressing.
- Resize and upscale
- Target dimensions, or 2×/4× machine-learning enlargement.
- Batch behaviour
- How a folder of files is processed, queued, and reported.
- Job control
- Plan, run, monitor, and cancel jobs mid-flight.
Local and private
MCP connects to the app on your Mac — not to a cloud service.
This is the structural difference. An AI client connects to Smol running locally, and every file is processed on-device from start to finish. Assets are never handed to a third-party processing service, never queued on someone else’s hardware, and never left behind in a cache. That makes the workflow usable for client work, unreleased product assets, medical and legal material, and anything under NDA — the categories that rule out an online tool entirely.
Setup
Connected in about two minutes.
- STEP 01
Open the Smol for AI area
Launch Smol on your Mac and select “Smol for AI” in the sidebar. The panel is titled “Connect AI tools to Smol”.
- STEP 02
Choose your AI client
Follow the guided setup for Claude Code or Codex. Each client shows a live status and reads “Connected” once the MCP connection is active.
- STEP 03
Copy the full AI guide
Use the “Copy full AI guide” button and paste the guide into your agent, so it knows every available tool, setting, and quality mode.
- STEP 04
Give your agent a media task
Ask it to compress, convert, upscale, resize, or strip metadata. The job runs locally on your Mac and the agent reads back structured results it can verify.
In practice
Six jobs an agent can now finish on its own.
Optimize website assets in-flow
An agent building or refactoring a site can compress and convert every image it touches to AVIF or WebP at a set quality target, without the developer leaving the editor or hand-running a build step.
Prepare app-store and marketing imagery
Produce correctly sized, correctly formatted screenshots and store artwork from a single source render, with repeatable settings across every locale and device class.
Convert design exports
Turn heavyweight exports from design tools into production-ready assets — format, dimensions, and weight all specified rather than guessed.
Strip sensitive metadata
Remove EXIF, GPS coordinates, author fields, and timestamps from user-supplied or client-supplied media before it is committed to a repository or published.
Compress release videos
Reduce demo recordings, changelog clips, and release footage with H.265 so they fit a documentation site or release page without a separate encoding tool.
Process large batches
Hand an agent a folder of hundreds or thousands of mixed files and let it apply one consistent standard, reading back per-file results as it goes.
More reliable AI workflows.
The practical gain is not that an agent can compress a file — it is that the agent can verify it did. Smol returns structured results, so the model sees the input size, the output size, the savings, and the paths it wrote, and can act on those numbers instead of assuming success.
Because settings are explicit and repeatable, the same instruction produces the same result on every run. An agent can apply a single standard across an entire project — one format, one quality target, one metadata policy — rather than making a fresh judgement call per file. And because it all happens through the tool layer, the developer never leaves the editor to run a separate utility.
Smol remains a standalone Mac app: drag-and-drop, Finder Quick Actions, a menu bar tray, and clipboard compression all work exactly as before. The MCP layer is additive. It expands who can use Smol’s tools, and how precisely those tools can be automated.
Available now
Native on Apple Silicon and Intel.
MCP support is included in Smol at no extra cost. Smol is a one-time $29 purchase with future updates included — no subscription, no per-seat AI add-on, and no usage metering on the agent tools.
Questions
Smol, MCP, and AI agents.
What is MCP, and what does Smol do with it?
MCP — the Model Context Protocol — is an open protocol that lets AI assistants and coding agents call external tools. Smol implements an MCP server, so an AI client can invoke Smol’s media-processing capabilities directly: compression, conversion, AI upscaling, resizing, and metadata removal for images, video, PDFs, and audio. The agent connects to the Smol app running on your Mac.
Which AI tools work with Smol?
Smol supports Claude Code and Codex out of the box, with guided setup from the in-app “Smol for AI” area and a live Connected status for each client. Because Smol implements the open Model Context Protocol, other MCP-compatible clients can connect using the same mechanism.
Can an AI agent do more than just “compress” a file with Smol?
Yes. Agents can run four distinct tool families — Classic compression, Neural detail-first image compression, Convert across images/documents/video/audio/tables/fonts, and Enhance for AI upscaling, metadata stripping, resizing, and batch automation. Within those, an agent chooses the output format and codec, quality target, compression mode, output location and naming, metadata removal, resize and upscale settings, and batch behaviour. It can also plan, run, monitor, and cancel jobs.
Are my files uploaded anywhere when an AI agent uses Smol?
No. MCP connects your AI client to the Smol app on your Mac rather than to a cloud processing service. Files are processed on-device and stay on-device for the whole job. That makes it viable for client work, unreleased product assets, and material under NDA that could not be put through an online tool.
Why route media work through Smol instead of letting the agent do it?
General-purpose agents either shell out to whatever encoder happens to be installed or guess at settings, and output drifts between runs. Smol gives AI tools a dedicated local finishing layer: production-grade encoders, Neural detail retention, explicit and repeatable settings, and structured results the agent can inspect and verify. The same instruction produces the same output every time, so one standard can be applied across an entire project.
Does the MCP layer replace the Smol app?
No. Smol remains a standalone Mac app with drag-and-drop, Finder Quick Actions, a menu bar tray, and clipboard compression. The MCP layer is additive — it expands who can use Smol’s tools and how precisely those tools can be automated.
Does this work on both Apple Silicon and Intel Macs?
Yes. Smol ships separate native builds for Apple Silicon and 64-bit Intel Macs, and the MCP capabilities are available on both. Apple Silicon requires macOS 13 or later; Intel requires macOS 14 or later.
How much does it cost?
Smol is a one-time $29 purchase that includes future updates. There is no subscription, no per-seat AI add-on, and no usage metering on the MCP tools.
About
About Smol
Smol makes powerful file compression feel like a native part of the Mac. It combines professional encoders and ML-guided image analysis with a simple drag-and-drop interface — and now an MCP tool layer for AI agents — keeping processing private, local, and unlimited. Smol is a one-time $29 purchase and runs natively on Apple Silicon and Intel Macs.
Media contact
Press and partnerships
For interviews, product information, review access, screenshots, or partnership inquiries, contact [email protected].
Previous release: Smol launches for Apple Silicon and Intel Macs