Can Claude generate images? No, but it can call a tool

No. Anthropic's docs say Claude is an image understanding model only: it can interpret and analyze images, but it cannot generate, produce, edit, manipulate, or create them. To get a new image in a Claude chat or in Claude Code, connect an image-generation tool over MCP: Claude writes the request, the tool renders the image, and Claude hands back its link.

Claude's side comes from Anthropic's Vision and Models overview docs and its Help Center article on custom connectors. Sume's side comes from MCP tools and gates and the Image API docs. All were read on 2026-09-27. Sume has no official connector for Claude: this is a custom connector pointed at Sume's hosted MCP server.

Why can't Claude make images?

Because its output is text. Anthropic's models overview says all current Claude models support text and image input and text output. An image you attach is input that Claude can interpret and analyze; nothing in that list produces an image file. Anthropic answers the question directly in its vision docs: Claude is an image understanding model only.

How do I get images in a Claude chat?

Add an image tool through MCP. Sume's hosted MCP server at https://mcp.sume.com/mcp has one, generate_image. Sume's basics page says hosted MCP still works but is not the primary path today: it is there for clients that speak remote MCP, such as Claude, while a backend calls POST /v1/images directly.

• In Claude, Cowork, or Claude Desktop, add it as a custom connector; Free users are limited to one. Add Sume to Claude as a custom connector has the steps.
• In Claude Code, run claude mcp add --transport http sume https://mcp.sume.com/mcp, then claude mcp login sume. Generate images for a website with Claude Code shows a full session.
• Either way, turn Write on at Sume's consent page. The default sign-in is read-only (mcp:read), and paid tools such as generate_image return insufficient_scope until the session also has mcp:write.

What happens when Claude calls generate_image?

Claude runs a short sequence of tool calls. In current code, generate_image submits to POST /v1/images and answers with a job id; stills typically take 10–60 seconds.

Can Claude edit an image this way?

Claude still does not edit pixels; the image model does. Sume's image API takes input_references for image-to-image and edit calls, and Claude can pass them to generate_image. MCP server for image editing goes further into edits. Two rules decide whether a reference works:

• Reference URLs must be public HTTPS. Localhost, private-network, and non-HTTPS URLs are rejected before submission, and hosted MCP cannot read files from your laptop.
• Models whose input_references descriptor is {"min": 0, "max": 0} are text-to-image only and reject references, so check the catalog entry first.

How much does it cost?

generate_image is a paid tool billed to your Sume wallet, and Sume's image docs say billing is all-or-nothing: a generation is either completed and billed in full, or it fails and is not billed. Each call needs an idempotency_key, dry_run=true shows the cost before anything runs, and max_spend_usd caps a call when you pass it. GET /v1/images/models lists the models with their pricing.