Your agent says what to make.
Connect AnyMake over MCP and your coding agent asks for an outcome, a product ad, a launch kit, a thumbnail, and gets back files in the right sizes. It never chooses a model.
What your agent gets
make_product_media: a product ad or product photos from one photo.make_launch_kit: a launch kit from a description, a screenshot and a logo.make_thumbnail: a thumbnail from a topic and the words to put on it.make_character_clip: a clip of a character in a scene.make: a goal in words, when none of those fits. AnyMake picks the recipe.auto_pickandwait_make: the Make takes the first candidate, and your agent waits until it is done. Nobody has to watch.estimate: true: on any outcome tool, the plan and the price. Nothing runs and nothing is charged.generate_imageandgenerate_video: one picture or one clip.search_assetsandget_asset: find what was made in the Gallery and fetch it.
Connect
Claude Code:
claude mcp add --transport http anymake https://anymake.dev/mcp \
--header "Authorization: Bearer <your key>" --header "X-AnyMake-Source: claude-code"
Codex, in ~/.codex/config.toml:
[mcp_servers.anymake]
url = "https://anymake.dev/mcp"
http_headers = { "Authorization" = "Bearer <your key>", "X-AnyMake-Source" = "codex" }
Claude Desktop and Cowork take the same URL, https://anymake.dev/mcp, with streamable HTTP transport and the same two headers.
Get a key
Sign in, open Account and create an API key. The key starts with anymake_sk_ and is shown once. Every call is paid from your account's credits at the price shown, and a failed job costs nothing. See the prices.
REST
The same handlers answer under /v1 with the same bearer. POST /v1/make/estimate prices a Make without running it. Read the API reference.