Veida MCP

An MCP server that actually generates images, with no API key and no account.

Most image MCP servers are a wrapper around a key you have to go and buy first. This one talks to the anonymous tier of Veida — a client id in a header instead of a login — so it works the moment it is installed.

> generate a 16:9 cover image of a paper boat on a calm lake at sunrise

  → https://cdn.veida.ai/uploads/kie/image/….webp

Tools

| Tool | What it does | |---|---| | generate_image | Prompt → hosted image URL. Submits, polls, returns. 40–120 seconds. | | list_models | The model shelf, which ones run signed-out, and the page for each. | | browse_tools | Pages worth sending a person to — editor, upscaler, background remover. | | free_tier_limits | The exact ceilings, so an agent does not build something that always fails. |

Install

Claude Desktop / Claude Code

{
  "mcpServers": {
    "veida": { "command": "npx", "args": ["-y", "veida-mcp"] }
  }
}

Cursor / Cline / Windsurf

Same block, in that client's MCP config file. There is nothing else to set — no env, no key.

What the free tier is, measured

• The anonymous grant is 4 credits and one image costs 4 — so the server mints a fresh client id per image rather than reusing one.
• A ceiling of 30 credits per IP per day sits on top: roughly 7 images a day from one machine, however many ids it invents.
• Free output is 1K and watermarked.
• 🔴 Editing an existing image cannot run anonymously. The editing model costs 30 credits against a 4-credit grant, so an anonymous edit can never be paid for. The tool set says so rather than offering a call that always fails; for edits, sign in and use the image editor or the ChatGPT image editor.
• There is no video on this site, so no video tool.

Signing in removes the watermark, lifts the cap and opens the rest of the shelf. The current lineup is on the models the generator offers: GPT Image 2 when the picture has to contain readable text, Nano Banana 2 for instruction following, Seedream 5.0 Lite for colour and composition, and Qwen Image 3 for typography. Pricing lists what each tier costs.

Writing prompts that work

Name the light, the material and the composition. "A product photo of a mug" gives the model nothing; "matte black ceramic mug on pale oak, soft window light from the left, shallow depth of field" gives it a picture. Worked examples live in the prompt library, and the showcase is there if you want to judge output quality before wiring anything up.

The same engine is what the ChatGPT image generator page serves in a browser — this server is the headless door to it.

Develop

npm install
npm run build
node dist/index.js      # speaks MCP over stdio

Links

• veida.ai
• Image generator
• Upscaler · Background remover

MIT.