Hedra posts that its MCP server and CLI are live for agent visuals

On September 30, 2026, @hedra_labs posted that Hedra MCP and CLI are live. The note calls that pairing "the first step toward our platform for building, improving, and deploying vision models," and tells people to connect Claude Code, ChatGPT, Hermes, or their own agents to Hedra's image, video, and world models "right where you already work."

That platform sentence is Hedra's framing. Neither the post nor the current connector page says agents now train or deploy custom vision models. Both describe a narrower thing: an agent the user already has open can request visuals from Hedra.

The connector page is the concrete half of the announcement. It offers visual understanding plus video, image, music, and voice generation, using the caller's Hedra subscription or pay-as-you-go. The HTTP server is https://mcp.hedra.com. The page names Claude, Claude Code, Codex, Cursor, ChatGPT, and Hermes. "World models" is the post's wording. The page's own list is visual understanding plus image, video, music, and voice.

For Claude Code, the documented install is one command and a browser sign-in:

claude mcp add --transport http hedra https://mcp.hedra.com

The page then says to run /mcp, complete Hedra sign-in in the browser window that opens, and ask in the next message. The sample prompt is a 16:9 hero image. On that path, the page says Hedra hands the finished file back to the session.

The same URL is documented for Codex, Cursor, ChatGPT, and Hermes. Codex and Cursor have their own add steps; ChatGPT is a custom app, and plans that gate it need developer mode first. Hermes stores the server in ~/.hermes/config.yaml with OAuth and caches the token after the callback.

Listed uses are generation and visualization: site and app art, campaign images, video from a sentence or starting image, and renders of a building or product before it exists. A gallery on the page claims one integration for top image, video, and speech models. The September 30 post does not name those models.

The page treats MCP as the fit for an agent you already use, and the API as the fit when a product owns parameters, sequencing, retries, and customer experience. A CLI link sits beside MCP, matching the post's "MCP + CLI are live," but this page does not document CLI commands.

The post is an availability note. It does not establish that the MCP endpoint appeared for the first time on September 30. What the post and the docs do establish is a public HTTP connector, a sign-in step, and a stated scope for seeing and generating inside the agent session.

If you want to try it, start on the connector page: add https://mcp.hedra.com in the client you already use, finish the Hedra sign-in, and run one image prompt before treating video, music, or voice as confirmed on your plan.

Sources