MCP — use Cosmiq from Claude
Connect Cosmiq to Claude (web, desktop, or Claude Code) and create & revise visuals by just describing them. Every request hands back a board link where the result lives, so you can refine, regenerate, or download it in Cosmiq.
Access to MCP / API is currently invite-only (private beta). Once enabled, connect as below.
Connect once
claude.ai (web) — Settings → Connectors → Add custom connector → URL
https://www.cosmiq.studio/api/mcp → authorize with your Cosmiq API key (csk_…).
Claude Desktop / Code — add the cosmiq server with your csk_ key, restart.
You never call tools by name — just describe what you want; Claude picks the tool.
What you can do
- One image — “Create an image of a serum on marble, warm editorial, 9:16.”
- A video — “Make a 5s launch video of these earbuds, 16:9, with audio.” (image → animated)
- A batch / campaign — “10 ad covers for [brand], 1:1” → one board of concepts (generate now or leave ready).
- A carousel — “Make an educational IG carousel about morning habits.” → one shared style + N slides, consistent look.
- Recreate a real product — give product photo URLs: “put each on a clean studio background, keep the product identical.”
- Place images — “put these image URLs on a board.” (no generation, no credits)
- Read & revise an existing board — paste a board link: “look at these slides and tighten the prompts.” Claude reads the board (sees the images), then rewrites the prompts in place.
What you get back
- board —
https://www.cosmiq.studio/boards/…(where your result lives) - for video — a direct video URL
The tools (under the hood)
| Tool | Does | Generates? |
|---|---|---|
cosmiq_create | One image, optionally animate to video | Yes (always) |
cosmiq_campaign | 1–12 concepts → one board | Optional (generate) |
cosmiq_carousel | Shared style + 2–15 slides → one board | Optional |
cosmiq_image_edit | Reference URL(s) + edit prompt → edit node | Optional |
cosmiq_add_image | Drop 1–20 image URLs onto a board | No (no credits) |
cosmiq_get_board | Read a board’s prompts + images (to revise) | No |
cosmiq_update_prompts | Rewrite prompt-node texts on a board | No |
Full parameters: API reference.
What you can NOT do (yet)
- Video only via
cosmiq_create. Campaign, carousel and edit produce images, not video. - No short-form video clipping (cut a podcast into Reels) via MCP.
- No publishing / scheduling to Instagram/TikTok/Buffer from MCP — you get a board to review and download; posting is done elsewhere.
- References need public URLs. Claude can analyze an attached image, but to wire it into a chain it needs a URL (or drop it onto the canvas in Cosmiq).
- On-image label text isn’t pixel-perfect (text-to-image limit). A reference keeps the product look, not always the micro-typography.
- Credits are charged to the API-key owner;
generate: trueon a 10-concept batch = 10× image credits. Building a board without generating is free. - Each create/campaign/carousel/edit/add call makes a new board; only
get_board/update_promptsact on an existing one.
Good to know
- Grouping — you can pass a name to group related boards under one workspace and reuse it next time. It is just organization, not a brand.
- Revise loop —
get_board(reads prompts + shows the images) → you decide changes →update_promptswrites them back; with live-sync the canvas updates as it happens.