blog/developers·Sep 28, 2026·3 min·by the eroq team
Generate images and video from Claude Code, Cursor and VS Code
Give your coding agent eroq's image, video and voice tools — one command or one click, a browser sign-in, no key in your config. Setup for Claude Code, Cursor and VS Code.
Every product needs pictures before it has a designer: the hero image of a landing page, empty states, avatars for a seed database, a demo clip for the README, a voice line for an onboarding flow. Your coding agent already knows where those files go. With eroq's MCP server connected, it can also make them — same models, same credit prices as the studio, and no API key sitting in a config file.
Claude Code
One command:
claude mcp add --transport http eroq https://eroq.ai/mcp
Add --scope user to make eroq available in every project. Then, in a Claude Code session, type /mcp, select eroq and choose Authenticate: eroq opens in your browser, you sign in and click Allow, and Claude Code stores its own key — listed in your dashboard as Claude Code (connected).
Prefer a plugin? eroq serves a Claude Code marketplace from its own site:
/plugin marketplace add https://eroq.ai/claude-code/marketplace.json
/plugin install eroq@eroq
The plugin declares the same server plus a skill that teaches Claude Code how eroq works — check list_models before spending, poll video instead of assuming it finished, download signed links before they expire — so you do not have to explain it in every session.
Cursor
Press Add to Cursor on eroq's Cursor page and confirm. The server lands in your MCP settings marked « Needs login »; click it, sign in to eroq, Allow. The equivalent ~/.cursor/mcp.json entry is just the URL:
{ "mcpServers": { "eroq": { "url": "https://eroq.ai/mcp" } } }
VS Code
Install in VS Code on eroq's VS Code page registers the server; start it when VS Code asks and sign in in the browser. It shows up in Copilot Chat's agent mode tools. The workspace version:
{ "servers": { "eroq": { "type": "http", "url": "https://eroq.ai/mcp" } } }
What to ask for
Agents are good at turning a vague request into a well-formed call. A few that work well:
Generate a 16:9 hero image for the pricing page: a calm desk at dawn, soft
window light, lots of negative space on the left for the headline. Save it
to public/images/pricing-hero.png.
Make 8 square avatars for the seed users in db/seed.ts — varied adults,
neutral studio background, photographic. Name the files after the usernames.
Record the onboarding line "Welcome back — your drafts are where you left
them" with a warm voice, and put the MP3 in assets/audio/.
The picture comes back to the agent with a link; it downloads the file into your repo. Links are signed and expire after 30 minutes, which is why the plugin's skill tells Claude Code to save immediately — every render also stays in your eroq library.
For work that is all files and no chat, the eroq CLI runs a local MCP server that writes renders straight to disk: npx -y @eroq/cli mcp (it reads EROQ_API_KEY). Codex CLI uses it that way.
What it costs
The studio's prices, from the same wallet:
- 10 credits per image on Krea 2 (20 on Eroq One, which holds up to ten reference photos).
- 45 credits for a five-second Seedance 2.0 Mini clip.
- 3 credits per 100 characters of speech on Voice One.
enhance_prompt,list_models, the status checks and every lookup: free.
Failed renders refund themselves. If an agent loops on generations, per-key rate limits and your workspace's monthly credit cap (set per member in the dashboard) are the brakes.
Keys and security
- Each client gets its own key through the sign-in, named after the app, revocable in Developers → API keys. Nothing to commit by accident.
- Reconnecting the same app replaces its key rather than adding one.
- Scripts that need a static key can still send
Authorization: Bearer eroq_sk_…— create a dedicated key for them.
FAQ
Does the agent see the images it generates?
Yes — generate_image returns the picture itself along with the link, so the agent can check it matches the brief before moving on.
Can it generate video for a README demo?
It can start the clip; renders take one to five minutes, so the agent polls get_video_status and downloads the MP4 when it lands.
Is this different from the Claude and ChatGPT connectors?
Same server and tools. The difference is where the output goes: in a coding agent it becomes files in your project.
All the one-click buttons are on eroq.ai/mcp.
Make this with the models behind the post — start with 50 free credits , or browse every engine and its price .