Before USB-Cs became standard, you needed one charger for your tablet, one for your kindle, and another for your headphones. Did your Android phone die at your all-Apple friend’s house? Sorry, you’re going to have to live like it’s 1999… at least till you go home to your charger.

Model Context Protocol, or MCP, is the USB-C of AI tools. You don’t need a different integration to connect each AI tool to each app you use. With MCP, your AI assistants (like Claude or ChatGPT) can connect to outside tools (like Notion) and pull in the information they need. 

Connect Air’s MCP, and the AI tools your team already uses, like Claude, can find, preview, and organize your assets without anyone opening Air. It won’t edit images or generate a newsletter header. But it does make your life easier by handling the finding and filing.

In this piece, you’ll learn what Air’s MCP does (and doesn’t) do, and how to set your library up for success to make sure the agent is as useful as possible. You’ll also find four workflows you can use as inspiration for your team.

Air gives you a ready-to-use, official MCP that lets your AI agents interact with Air assets, boards, custom fields, tags, imports, and uploads. Here’s what you can actually use it to do on a day-to-day basis:

Find and preview assets

You’re looking for the approved hero images for your spring campaign. But there are 73 Air boards to dig through and you can’t find your assets. 

If you’ve connected Air to Claude, you could use this prompt: 

"List every asset on the Spring Campaign board tagged 'hero' where approval status is Approved, and show me previews."

Your AI agents can search the library and narrow results by Board, Tag, or Custom Field value. Then, they preview what they find so they can confirm the right asset before showing it to you. 

Everything, from the prompting to the results, happens inside Claude, so you don’t have to leave the tab at all.

[Image suggestion: Trying this prompt, or similar, in Claude and showing the results.]

This isn’t the same as using the plain-English Conversational Search tool inside the Air app. This is a structured retrieval running against existing Air search and filters. Your AI agent works best when the request names a board and a filter rather than describing a vibe. But more on this later.

Retrieve download and CDN links

If your developer needs the latest logo for the website, Claude can grab the current version’s download URL or CDN link. 

[Image suggestion: Trying this prompt in action, picture of the results]

But how do you know it’s actually the current final version and not the “asset_final_final_Iswearitsalmostfinal” version?

Air’s Version Stacking keeps every iteration together with the live approved cut on top. A current-version CDN link points to whichever version is on top, instead of a stale export sitting in someone’s Downloads folder. Add a new version in Air and the CDN link follows it wherever it’s live.

Write metadata back to the library

Yikes, your team just renamed a product line. Don’t worry. You don’t need to cancel your after-work gym session just to rename every asset tagged with the old name.

Agents can rename assets, apply and remove tags, and set custom field values. 

This means campaigns, channels, product lines, usage rights, and approval statuses all become fields an agent can read and populate. Ask it to fill a blank field across a set, and it writes the values back in one pass.

Create boards and file new assets

Every creative team has someone who spends half of the afternoon downloading files from email, uploading them to the library, renaming them, dropping them into the right folder, and making sure the new version replaces the old one. 

It’s important work, and someone’s gotta do it. The MCP makes sure that someone isn’t you…or any other poor human on your team.

AI agents can create boards and file assets into them, import a file from a URL, or upload a local file directly. A single instruction can bring in a file and drop it into the right board.

MCP is a powerful protocol, but it’s not built for certain tasks. Don’t worry though. We’ll also tell you which tool you can use for what MCP doesn’t cover.

It doesn’t edit, resize, or generate

MCP lets AI agents find, organize, and upload your assets. It doesn’t do image editing, reformatting, background removal, or generation.

Need to crop a hero shot for TikTok? Use Air Canvas for editing and reformatting. Inside Canvas, you can remove and extend backgrounds, upscale the resolution, resize by channel, and more. 

Context, Air’s workspace-level brand memory layer, keeps brand colors, fonts, logos, and usage rules applied to every edit.

It isn’t Agentic Chat

Air’s MCP isn’t the same thing as Agentic Chat, Air’s own natural-language interface. 

You can use Agentic Chat to describe edits, resizes, or bulk adaptations and the brand kit and creative library are already built in. MCP runs the opposite direction. When you connect AI agents to your creative asset library, it opens the library to tools that live outside Air.

This impacts how specific your prompts have to be. 

Agentic Chat can take a loose, descriptive request because it sits inside Air’s context. An outside agent using MCP gets sharper results when the request names boards, tags, or field values.

It doesn’t go around your Air permissions

Air hosts the MCP server, so there’s nothing to install or run on your own machine. You add Air’s MCP URL to your AI client as a connector, and the connection runs through your own Air account. 

Access is also scoped to the credentialed user’s permissions. AI can’t access something you yourself wouldn’t be allowed to access.

Underneath any agent access sit Air’s broader governance controls. On Enterprise plans, these include:

• Single sign-on (SSO) through SAML 

• Custom roles and permissions 

Learn more about how we protect your data in Air’s security guide.

Don’t ask what an MCP can do for you, ask what you can do with an MCP. Here are four practical workflows you can copy:

1. Pulling the approved hero set for a campaign brief

The marketing lead pastes the brief into Claude. The agent queries Air for assets on the relevant board, filtered by approval status and channel. It previews the matches and returns download links for the ones that fit the brief.

Compare that to the manual version. You ping a designer, wait for a reply, then receive a Drive folder that may or may not hold the current cut. Half the time you’re back to asking which file is final.

Using the MCP lets you start your campaign right from approved work.

2. Auditing assets missing a usage-rights field

As you know, a single asset used past its license window can lead to real legal consequences — especially if you’re in sports, media, or retail.

Unfortunately, this is the kind of cleanup that rarely gets prioritized. Nobody volunteers to click through hundreds of assets by hand.

Use your agent to query for assets where the usage-rights custom field is empty. Once your ops lead supplies the licensing terms, the agent can write the values back across the whole set in one pass.

3. Importing a shoot from a URL into the right board with tags

A photographer delivers via a hosted link. The agent can import from that URL, select the destination board, and apply campaign and product tags on the way in. The shoot lands where it belongs, tagged, without a manual pass.

Air’s AI engine then auto-tags those uploads by concepts, colors, and objects on top of whatever the agent applied. The assets arrive searchable, layered with both the agent’s structured tags and Air’s visual recognition.

This way, intake stops relying on one person’s folder discipline and their memory of where things go.

4. Grabbing current-version CDN links for a build

Want the right image in your code without the download-upload shuffle?

Inside Cursor or Windsurf, the agent can retrieve CDN links for the approved assets a page needs and drop them straight into the code.

When a new version is added in Air, a current-version CDN link follows it wherever it’s live, so the site picks up the new version without redeploying.

An agent is only as good as the library it queries. If you’re coming from scattered Drive and Dropbox folders, you need a system of record first. 

It’s an upfront time investment, but it saves you from some major headaches later on. Here’s how to go about it:

One authoritative library instead of five

Even the smartest AI can struggle to track down what you need if it has to comb through Drive, Dropbox, Box, SharePoint, local machines, and a legacy Digital Asset management (DAM) system.  It can take a fair bit of time and consumes more usage credits. Plus, the results may not be as clean.

Air closes that gap two ways. 

1. Cloud Imports migrate archives directly from Google Drive, Dropbox, Box, and SharePoint into the workspace.

2. Air Flow, Air’s Mac desktop app, syncs your library into Finder and saves edits back to Air as new versions, so work in progress doesn’t live outside the system. 

Now you have a single library that both people and agents address, instead of five that neither can search cleanly.

Metadata an agent can actually filter on

The MCP searches by boards, tags, and custom fields, so your results are only as good as how well those are set up. 

Use this checklist to tidy things up:

1. Consistent board structure across campaigns and teams

2. Campaign and channel captured as custom fields

3. Approval status recorded as a field, not a note

4. Usage rights and expiration dates filled in

5. Version history intact, with the current cut on top

But don’t worry, the task isn’t as daunting as it sounds. Air cuts down on a lot of manual tagging. It: 

• Auto-tags by concept, color, and object

• Uses facial recognition to organize content by talent

Much of the metadata builds itself as assets come in. 

Guardrails that keep agents off unapproved work

Without the right guardrails, an agent that can retrieve anything will eventually retrieve a work-in-progress file, an expired asset, or an off-brand version.

Air has a few controls in place to prevent that:

• Approval status as a custom field the agent filters on

• Access-controlled Libraries (Enterprise) to separate brands or regions

On your end, make sure a human’s actually approving everything before you ship.

A real DAM with AI doesn’t mean a smarter search bar bolted on. It means AI agents can query your library directly via MCP, instead of waiting for someone to export files into a chat. 

You can open it up to the AI tools your team already uses. This way, the AI can do the finding, filing, and organizing that eats up your people’s time. And since the AI is working from the same library your team uses, it can reach for the approved, current version of your work when your fields mark which one that is.

You can get there in three steps:

1. Consolidate everything into one system of record.

2. Get metadata and approval status clean and consistent.

3. Connect the agents your team already uses through MCP.

This stops the library from being a place people dig through. It becomes infrastructure that both your team and your agents can rely on.

Ready to try it? Follow Air’s MCP setup guide and check Air’s pricing page for plan availability. Then start a workspace.

What is an MCP server for digital asset management?

An MCP server for digital asset management is a connector that lets AI agents query your DAM and act on it. For Air, it means agents can find assets, retrieve links, update metadata, create boards, and import files directly, working from the live library, not static exports.

Where does Air MCP run and which clients are supported?

Air hosts its MCP server, so there’s nothing to install on your own machine. Air’s developer docs name these AI clients:

• Claude

• Cursor

• Windsurf

• ChatGPT

To connect Claude, open Settings → Connectors, choose Add custom connector, and paste Air’s MCP URL: https://mcp.air.inc/mcp. Air’s developer docs point Cursor, Windsurf, and ChatGPT users to the same setup guide.

On the Claude side, you’ll need admin access to your organization’s Claude subscription, or your own personal Claude subscription.

Do I need a developer to set up Air’s MCP server?

No. In Claude, setup means adding Air’s MCP URL as a custom connector in Settings. On a company Claude plan, that takes admin access to the Claude subscription. Once it’s connected, an ops or brand lead uses it entirely through plain prompts in their AI client, no code required.

Can AI agents edit or resize assets through MCP?

No. MCP handles retrieval, organization, metadata, imports, and link generation, not editing, resizing, background removal, or generation. Those tasks happen in Canvas, where Smart resize handles channel sizing and Context supplies your brand details for each edit.

How does Air keep agents from retrieving unapproved or expired assets?

Access is scoped to the connected user’s permissions. Plus, you can track approval status in a custom field. This way, the AI can filter to the approved materials only. You can also add fields like expiration date, so agents don’t pull expired assets.

On Enterprise plans, access-controlled Libraries also keep brands or regions apart. Either way, have a human check anything headed for external distribution.

What’s the difference between Air’s Agentic Chat and MCP?

Agentic Chat is Air’s own natural-language interface inside the app, where the brand kit and library are already loaded for edits and adaptations. MCP runs the opposite way, opening the library to outside agents like Claude or Cursor so they can query and act on it.

Which Air plans include MCP and API access?

The Business and Enterprise plans include public API and MCP server access. Confirm current availability on Air’s pricing page.