Content
<a href="https://gola.apimart.ai/k-EXCk8v"><img src="banner.jpg" alt="GPT Image 2.6" width="100%" /></a>
# GPT Image 2.6 MCP Server — use it from Claude, Cursor, VS Code
> Expose GPT Image 2.6 as MCP tools so agents can generate images on their own.
[](https://github.com/sindresorhus/awesome) [](LICENSE)
**[👉 Try GPT Image 2.6 on APIMart](https://gola.apimart.ai/k-EXCk8v)**
`https://gola.apimart.ai/k-EXCk8v`
[💰 Pricing](https://gola.apimart.ai/k-74p0Xy) · [🔑 Get an API key](https://gola.apimart.ai/k-FpTbac)
---
## What this is
An MCP (Model Context Protocol) server that exposes GPT Image 2.6 to Claude
Desktop, Claude Code, Cursor and any other MCP-compatible client.
```jsonc
// claude_desktop_config.json
{
"mcpServers": {
"apimart-image": {
"command": "npx",
"args": ["-y", "@apimart/mcp-image"],
"env": { "APIMART_API_KEY": "YOUR_KEY" }
}
}
}
```
## Tools exposed
| Tool | Arguments | Returns |
|---|---|---|
| `generate_image` | `prompt`, `model?`, `size?` | image URL + cost |
| `list_models` | — | available image models |
| `get_task` | `task_id` | status + result |
## Example session
```
you: draw a minimal poster for a coffee shop called "Morning Ritual"
claude → generate_image(prompt="minimal poster, serif headline 'Morning Ritual' ...")
← https://.../out.jpg (cost $0.0198)
```
## Why MCP instead of a script
- The agent decides when to call the model, no copy-paste loop
- Cost is returned per call, so the agent can reason about budget
- Works across clients (Claude, Cursor, VS Code) with one config
## Build it yourself
The server is ~120 lines: one MCP tool definition per endpoint, plus polling.
See `docs/` for the reference implementation.
---
## Related
- Prompt library: [gpt-image-2-6-prompts](https://github.com/fujc683/awesome-gpt-image-2-6-prompts)
- More image models: [awesome prompt collections](https://github.com/fujc683?tab=repositories)
## License
CC BY 4.0