Can Claude Code list items for sale from a folder of photos?

Yes. Run the ClearList MCP server locally, point Claude Code at a directory of photos, and it reads them, groups them by item, generates a listing for each, and publishes them to one sale page. Batches run up to 50 photos per call, so a few hundred photos is several calls rather than several hundred.

How is this different from using ClearList in the Claude app?

The app is one conversation and you upload photos by hand, which works well up to roughly thirty items. Claude Code reads the folder itself, so the item count stops mattering and the run is unattended. The tradeoff is that you need a terminal and a local install, so it is the wrong tool for someone selling a couch.

Can I use ClearList with Gemini?

Not in the Gemini chat app, which does not accept custom MCP servers. You can use it through Antigravity, Google's agentic IDE, which does: add ClearList to its mcp_config.json and it works the same way as any other stdio MCP server. So the Google path to ClearList is the folder workflow rather than a chat conversation.

Do I need an API key to run ClearList locally?

You need one, but you do not have to go and fetch it. Over the local stdio transport, account creation is itself a tool: the agent asks for your email, ClearList sends a 6-digit code, you paste it back, and the agent receives a key it stores itself. No browser, no dashboard visit. This is the one path where that fully in-conversation bootstrap works; the hosted endpoint uses OAuth instead.

How many photos can I process at once?

Up to 50 per call. The AI groups them by item automatically, detecting when several photos are different angles of the same thing, so you do not have to sort them into folders yourself. For a whole house you make several calls, which an agent handles without being asked.

Point Claude Code at a Folder of Photos and Sell a Whole House

There is a point in clearing out a house where the tool stops being the bottleneck and the uploading does.

I have written about using ClearList inside the Claude app, and for most people that is the right version. You photograph a thing, drop it in the chat, and a listing comes back. It holds up nicely to about thirty items.

Past that, something silly starts happening. You are not thinking about prices or descriptions any more. You are managing a queue of image uploads, a handful per message, keeping track of which ones you have already sent. The AI is doing its job. You have become a file transfer protocol.

The folder workflow exists for that.

What it looks like

You have a directory. Two hundred photos from walking around the house with your phone, unsorted, several angles of most things, some blurry ones you meant to delete.

You point an agent at the directory and say list everything in here.

Then you go do something else.

It reads the files, sends them up in batches, and the grouping happens server side: the AI works out which photos are the same object from different angles, so you do not have to sort them yourself. Each detected item gets a title, description, price, dimensions, weight, condition, and a flag for whether it needs a truck. Then it publishes to one sale page and hands you a link.

The item count stops being a number you think about. That is the whole difference.

Setting it up

This runs on the local transport, not the hosted connector. Install the server:

npm install -g @clearlist/mcp-server

Then register it with your agent. In Claude Code:

claude mcp add clearlist -- npx -y @clearlist/mcp-server

Set CLEARLIST_API_URL to https://clearlist.me. If you already have a key, set CLEARLIST_API_KEY too. If you do not, leave it out and keep reading.

The part where you do not go and get a key

Over the local transport, creating your account is itself one of the tools.

The agent calls send_verification_code with your email address. A 6-digit code lands in your inbox. You paste it back, the agent calls verify_code, and it receives a key that it stores on its own. You never open a browser, never visit a settings page, never copy a 68-character string out of a dashboard and into a config file.

Worth being precise, since the two transports differ here: this fully in-conversation bootstrap is a stdio thing. The hosted endpoint at /api/mcp requires OAuth and sends you through a consent screen instead. Both are passwordless. Only this one skips the browser entirely.

Does this work with Cowork, Codex, and Antigravity?

The pattern does, yes. Nothing here is Claude Code specific.

The requirements are that the agent can read local files and speak MCP over stdio. Claude Code, Cowork, Codex, and Google's Antigravity all qualify, and the registration syntax is the only real difference between them. Install the server, supply credentials, point at a directory.

Antigravity is worth calling out separately, because it is the only way to reach ClearList from inside Google's tooling. The Gemini chat app does not accept custom MCP servers. Antigravity does: you add the server to mcp_config.json, either globally at ~/.gemini/config/mcp_config.json or per workspace under .agents/mcp_config.json, and it behaves like any other stdio server.

So if you are a Google user wondering why there is no "sell your stuff with Gemini" guide, this is the answer. The Google path runs through the folder workflow, not through chat.

If you are choosing between them for this particular job, the thing that matters is not the model. It is whether the tool will sit there and work through two hundred files without asking you to confirm every batch.

Where the limits actually are

Fifty photos per call. A house is therefore several calls, which an agent chains without being told, but it is worth knowing the number exists rather than discovering it at photo 51.

The grouping is good and it is not perfect. Two similar dining chairs sold separately can come back as one listing with several angles. Skim the results before publishing. This takes a couple of minutes and it is the only manual step that genuinely earns its time.

Blurry and dark photos produce worse listings, same as anywhere. The agent will not tell you a photo was bad, it will just write a vaguer description from it.

And the thing no version of this does: it cannot give a buyer your address. There is no tool for it and no API path that reaches it with agent credentials, on any transport. That one stays with you, in the app, on purpose.

Who this is actually for

Narrow audience, and I would rather say so than pretend otherwise.

You need a terminal, an agent already installed, and enough stuff that batch processing beats doing it by hand. Someone selling a couch should not read this post. They should open the app, take one photo, and be done in four minutes.

But if you are clearing an estate, emptying a storage unit, or moving out of a house you have lived in for a decade, the arithmetic changes completely. Two hundred photos through a chat window is an evening. Two hundred photos through a folder is a command and a cup of coffee.

The listings are identical either way. It is the same engine. You are just choosing whether you or the agent does the carrying.