Gumnut + MCP = Agentic Photos

One of the developments I’ve been most excited about over the past year is the rise of agentic AI — AI that can not only chat with you, but also take actions on your behalf.  Many of the painful use cases around curation of photos and creation of digital and physical goods from photos become straightforward when AI can access your photos and take actions on your behalf.

In my first newsletter article, I described some of the ideas that came up during the conversations I had with people and businesses about their photo organization challenges:

All of these were possible with Gumnut’s APIs and SDKs before — but they required an understanding of how to write software.

So over the past month, I added an agentic chat interface to Gumnut, as well as the ability to securely connect industry-leading agents like Anthropic’s Claude to your Gumnut photos.  With the power of Gumnut combined with the intelligence of agentic AI, these use cases are now as easy as talking to a chatbot!

 In today’s post, I’m going to show you some basic examples of what you can do with Gumnut and AI agents.

Gumnut in Claude

Most people have heard of ChatGPT; Anthropic’s Claude is similar in that you can chat with it, but you can also connect a variety of tools to Claude to enable it to fetch information you have access to and take actions on your behalf. 

In Claude’s Settings, you can securely connect your Gumnut photos by adding Gumnut as a new Claude Connector.  This works in both Claude on the web, as well as Claude Desktop.  Just click “Add custom connector” and use https://mcp.gumnut.ai/mcp as the “Remote MCP server URL”.

When you add Gumnut as a connector, Claude will ask you to authorize it to access your photos.  It looks something like like this:

After you approve access, you can ask Claude to do things with your photos.  For example, Claude can help you look through your photos for great pictures of your family to include in a holiday card.

Here’s another example of using using Claude to curate photos from your photo library:

Claude worked at this for a couple of minutes, so I’m not going to include the full chat transcript, but here’s what it ended with:

Finally, here’s an example of Claude looking at a photo and describing it — note that the photo has no caption or other description in advance!  Claude is doing this by looking at the photo itself, as well as the file metadata and the faces and people that Gumnut found in the photo.

Gumnut Embedded Chat

In addition to integrating with agents like Claude, I added an embedded chat interface to the Gumnut dashboard:

Note that the embedded chat knows the context of what you’re looking at!  If you’re familiar with AI agents for software development like GitHub Copilot or Cusor, it’s the same idea — but working with your photos library instead of a software project.

In general, I’ve found that the embedded chat interface in Gumnut is useful for working within a photo library, but standalone agentic chat interfaces like Claude are better for working across your photos and other tools.  For example, Claude can take metadata about your photos and put it in a spreadsheet, or plot your photos on a map.  Claude can also caption your photos for you, then send your family a postcard with the photos and its caption.

Gumnut in Claude Code

For software developers, or anyone comfortable with a terminal app, Gumnut can even be used in Claude Code.  This terminal-based tool is great for when you want to have more fine-grained control over what your agent is doing, or if you want to have multiple agents working on a task in parallel.

Here’s an example of using Gumnut in Claude Code.  Note how I can collaboratively go back and forth with Claude to accomplish more complex tasks:

> Use the photos mcp server to tell me what albums I have.

⏺ photos-mcp-server - list_albums (MCP)

⏺ You have 8 albums:

  - Summer Trip 2025 (created Aug 12)
  - Lady Gaga — Mayhem Ball (created Jul 29)
  - Spring Break (created Jun 19)
  - Hendy Woods 2025 (created Jun 17)
  - Calvin Camera (created Jun 9)
  - Family (created Jun 4)
  - Passport Photos (created Jun 4)
  - Random (created Jun 4)

> How many photos do I have in each album?

⏺ photos-mcp-server - list_albums_assets (MCP)

⏺ Here's the photo count for each album:

  - Summer Trip 2025: 2 photos
  - Lady Gaga — Mayhem Ball: 7 photos
  - Spring Break: 10 photos
  - Hendy Woods 2025: 16 photos
  - Calvin Camera: 7 photos
  - Family: 41 photos
  - Passport Photos: 5 photos
  - Random: 8 photos

  Total: 96 photos across all albums

> Look in the Hendy Woods album and tell me about the photos in that album.

⏺ photos-mcp-server - list_albums_assets (MCP)

⏺ The Hendy Woods album contains 16 photos taken between June 13-16, 2025. Here's what I can tell you about them:

  Location: All photos were taken at Hendy Woods State Park in Mendocino County, California (coordinates around 39.07-39.08°N, 123.47-123.48°W)

  Devices: Mix of Google Pixel phones:

  - Pixel 9 Pro (6 photos)
  - Pixel 7a (10 photos)

  Timeline: The photos span 3 days:

  - June 13: 2 evening photos (6-7 PM)
  - June 14: 8 photos throughout the day (7:45 AM - 3:02 PM)
  - June 15: 5 photos (morning to late evening)
  - June 16: 1 morning photo (10:20 AM)

  Photo types:

  - Several portrait mode shots (indicated by .PORTRAIT in filename)
  - Motion photos (indicated by .MP suffix)
  - All are JPEG images

  The photos appear to document a camping or hiking trip to Hendy Woods State Park, with shots taken at various locations within the park over a long weekend.

> What people show up in that album?

⏺ photos-mcp-server - list_people (MCP)

⏺ The Hendy Woods album includes photos of 5 people:

  - Ted
  - Ender
  - Calvin
  - Beckett
  - Pamela

  This appears to be a family camping trip to Hendy Woods State Park with all 5 people appearing in various photos throughout the 3-day weekend trip in June 2025.

> How many times does each person appear in that album?

⏺ photos-mcp-server - list_faces (MCP)

⏺ Here's how many times each person appears in the Hendy Woods album:

  - Ted: 2 photos
  - Ender: 3 photos
  - Calvin: 2 photos
  - Beckett: 3 photos
  - Pamela: 1 photo

  Some photos contain multiple people (like asset_kxztwkWnVhkUspDXEx85xG has both Ender and Calvin, and asset_ccwqwKHGaBH9XnMkWMVcLo has both Ender and Beckett), while other photos in the album don't have any detected faces.

> Create a new album called "Beckett at Hendy Woods" and add the photos of Beckett from the current "Hendy Woods" album to the new one.

⏺ I'll create a new album called "Beckett at Hendy Woods" and add the photos containing Beckett from the Hendy Woods album.

⏺ photos-mcp-server - create_albums (MCP)

⏺ photos-mcp-server - add_albums_assets (MCP)

⏺ Done! I've created the new album "Beckett at Hendy Woods" and added the 3 photos containing Beckett from the Hendy Woods album:

  - One from June 14 morning (8:17 AM)
  - One from June 15 afternoon (12:29 PM)
  - One from June 15 evening (7:44 PM)

  The new album now contains all the photos of Beckett from your Hendy Woods 2025 trip.

Endless Possibilities

The examples above are proofs of concept that I chose because they fit nicely into a screenshot, but longer, more complex workflows are absolutely possible.

To illustrate this, I asked Claude to tell me what it thought that it could do:

Try it out!

Gumnut is available to anyone that wants to try it out — just head over to app.gumnut.ai to get on the waitlist.  I’m currently limiting signups because I don’t have all of the resource usage controls I’d like to have in place, but it’s all totally usable, if a little unpolished.

Getting Involved

Building this all from scratch has been an exciting journey for me, but I love working with and talking to other people interested in the space.

• If you have a background in B2B Sales / GTM and a strong interest in the photos space, I’d be interested in working with you.  I’ve been focused on building the platform and use cases, but haven’t spent much time on potentially commercializing it.

• If you have a background in Applied AI as it relates to photos and videos, I’d love to chat with you.  While I believe my background and experience gives me the fairly unique ability to build a platform for Photo Intelligence from scratch, I’d love to learn more from people with deeper experience on the AI side.

• If you have personal or business use cases for Photo Intelligence, I’d love to hear about them, especially if I haven’t mentioned them in this newsletter in the past.

• If you’d like to try building something with the platform, I’d love to hear from you.  I’m not making anything self-serve right now, both because the platform isn’t ready for scaled consumption and because I want to understand everyone’s use cases better and work with them directly as needed.

In all of the cases above, drop me a comment below, or send me an email.