Can Claude make videos? Only by calling a video tool

Claude's output is text, not video. It can still get you a clip by calling a video tool, such as Sume's hosted MCP server added as a connector.

5 min readSume
All posts

Not by itself. Anthropic's docs list text and image input and text output for all current Claude models, and video is not on that list. Claude can still get you a video by calling a video-generation tool on a remote MCP server, such as Sume's: Claude writes the request, the server renders the clip, and Claude hands back its link.

Claude's side comes from Anthropic's Models overview and Vision docs, its Help Center article on custom connectors, and the Claude Code MCP docs. Sume's side comes from the MCP quickstart and MCP tools and gates. 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.

What can Claude do for a video without a tool?

Plan it in text. What Claude returns on its own is text, so it can draft the script, the shot list, or the prompt a video model needs, but not the video file itself. It does not draw the frames either: Anthropic calls Claude an image understanding model only (can Claude generate images?). The same models overview that lists text output also lists tool use, and a tool is how a conversation with Claude ends in a real clip.

What does a video tool add?

The rendering. Anthropic's article says remote MCP servers give Claude tools it can invoke during your conversation, and the server's developer defines what those tools do. Claude fills in the request; the server runs the video model. With Sume's hosted MCP server, a clip takes three steps:

  • Claude calls generate_video with your prompt and settings. In current code, the tool answers with a job id within milliseconds, never a clip URL.
  • Claude waits with jobs_wait. One call holds at most 55 seconds, so on wait_slice_expired Claude waits again on the same job; the docs say never to resubmit the paid create.
  • Claude reads the finished job with jobs_result. Sume returns generated outputs as Sume-hosted artifacts under media.sume.com, so the link Claude hands back is on that host.

Can Claude Code make videos too?

Yes, through the same server. In Claude Code, add Sume's hosted server over HTTP, then run claude mcp login, which runs the server's OAuth flow from your shell. In Claude, Cowork, or Claude Desktop, add https://mcp.sume.com/mcp as a custom connector instead; Add Sume to Claude as a custom connector has the steps.

Either way, turn Write on at Sume's consent page. The default sign-in is read-only (mcp:read), and paid tools return insufficient_scope until Write adds mcp:write. Connect Claude Code, Cursor, or Codex to Sume covers API-key sessions.

claude mcp add --transport http sume https://mcp.sume.com/mcp
claude mcp login sume

What kinds of video can Claude start through Sume?

Each of these is a paid tool that submits a job:

From MCP tools and gates, the Models overview, Video Generation, and Timeline 1.0, read 2026-09-27.
You ask forSume MCP toolNotes
A clip from a text promptgenerate_videoText-to-video through sume/auto or a named catalog model
A clip that starts from your imagegenerate_video with frame_imagesImage-to-video; the image must be at a public HTTPS URL, since hosted MCP cannot read files from your laptop
A clip guided by reference imagesgenerate_video with input_referencesReference-to-video
A presenter speaking your scripttts_create, then avatar-image-to-video_createVEED Fabric 1.0 turns a still and the speech into a talking clip; see UGC videos with Claude
One MP4 from several clips and an audio tracktimeline_createOne audio spine plus 1–200 video slots, every URL already on this workspace's media.sume.com

How long does it take, and what does it cost?

Video jobs run asynchronously. Sume's docs say video generation typically takes 30 seconds to several minutes, depending on the model and parameters.

Every tool in the table bills your Sume wallet. Ask Claude to call a tool with dry_run=true first: that previews the cost without submitting the job. max_spend_usd caps a call, and is enforced only when provided. For per-model prices, GET /v1/videos/models lists each video model with its pricing. Hosted MCP suits a chat like this one; Sume's basics page says it still works but is not the primary path today, so a product that makes video calls the Developer API or a Format.

Sources

Related posts

More in Agents

All Agents posts

Written by Sume