Skip to main content

Connect with MCP

Aether runs a remote Model Context Protocol (MCP) server. AI assistants that support MCP, such as Claude, Claude Code, Cursor, Antigravity, Codex and ChatGPT, can use it to generate images, effects and sprites and to tune images with your Aether account.

Server URL​

https://api.aetherforgeai.com/mcp

The server uses the Streamable HTTP transport.

Add the server URL to your client. The first time the client connects, your browser opens the Aether connection page.

  1. Log in to Aether if you are not logged in yet. You return to the connection page automatically.
  2. Check the client name and the address you will be sent back to.
  3. Select Connect.

The client then receives its own access token. Access tokens expire after one hour and the client renews them automatically.

Pick your client:

claude mcp add --transport http aether https://api.aetherforgeai.com/mcp

Then run /mcp in Claude Code and choose aether to start the login.

Connect with an API key​

Clients that can send custom headers can use an API key instead of logging in. Claude and ChatGPT connectors only support account login.

Authorization: Bearer <api_key>
claude mcp add --transport http aether https://api.aetherforgeai.com/mcp \
--header "Authorization: Bearer YOUR_API_KEY"

Tools​

Every tool uses the same options and credit costs as the matching page in the web app.

ToolWeb featureCredit cost
create_image_uploadUpload an image for other toolsFree
get_job_statusJob status and result URLsFree
get_creditCredit balanceFree
list_my_ai_modelsModels available for Aether generationFree
list_my_resultsHistory (tool, sprite, aether)Free
generate_imageGenerate image1 (GPT Image 2), 2 (GPT Image 2.5 High, Nano Banana Pro), 3 (GPT Image 2.5 XHigh)
generate_effect_v1Generate effects, V1 tab5 (1K), 7 (2K), 10 (4K)
generate_effect_v2Generate effects, V2 tab3
upscaleUpscale5 (1K), 7 (2K), 10 (4K)
inpaint_v1Inpaint, V1 tab5
inpaint_v2Inpaint, V2 tab3
remove_backgroundRemove BG1
equip_variationEquip variation5
sprite_liteSprite Lite5 to 13 by frame count
sprite_plusSprite Plus9 (49 frames), 13 (81 frames)
sprite_zenithSprite Zenith9 (36 frames), 25 (100 frames)
sprite_pulsarSprite Pulsar3, or 4 for Sprite To Sprite
aether_generateAetherFree for included models, otherwise 2. list_my_ai_models shows credit_cost per model

Field names and allowed values are the same as the matching endpoint pages.

Results​

Generation tools start an asynchronous job and return a job_id right away. Call get_job_status with that job_id to get the result. Set wait_seconds (up to 60) to wait for the job before the tool returns. When status is Succeed, image_urls holds the result image, and the GIF for sprites. See Asynchronous Job Lifecycle for the status values.

generate_image also creates a chat room on the web History page, listed under the API/MCP tab. The chat_room_id in the result identifies it, and you can keep generating in that room on the web.

Image inputs​

Image fields take one of these:

  • An upload_id from create_image_upload. Call the tool with the image's Content-Type, upload the file to the returned upload_url with an HTTP PUT within 5 minutes, then pass the upload_id.
  • The job_id of one of your finished jobs, to reuse its result image. This lets you chain tools, for example upscale and then remove_background.
curl -X PUT -H "Content-Type: image/png" --data-binary @character.png "<upload_url>"

Send exactly the Content-Type returned by the tool. Images can be up to 20MB. Uploading needs a client that can run commands or send HTTP requests, such as Claude Code, Cursor, Antigravity or Codex. In chat apps that cannot upload files, reuse a finished job_id instead.

Errors​

A failed tool call returns an error message instead of a result. When you do not have enough credits, the message links to the pricing page.