A Skill can organize prompts, run scripts, name files, and coordinate an image workflow. Whether it can call an external API directly depends on where the Skill runs: Claude Code has full network access, claude.ai varies by settings, and Claude API Skill containers have no network access.
A remote MCP connector provides a consistent way to expose image-generation tools to Claude.
Anthropic documents different runtime constraints for Claude Code, claude.ai, and Claude API Skills.
Claude Code Skills run with the same network access as other programs on your computer. A Skill can call an external API if its scripts and credentials are configured safely.
On claude.ai, a Skill may have full, partial, or no network access depending on the account and administrator's network-egress settings.
Skills used through the Claude API run in a code-execution container without internet access, so the Skill itself cannot call an external image API.
Claude Imagine's remote MCP server exposes generate_image, list_models, and get_credits without requiring the Skill to contain the image service itself.
Keep repeatable workflow instructions in the Skill and image generation in a purpose-built connector.
Decide whether the workflow will run in Claude Code, claude.ai, or through the Claude API because the network rules are different.
In Claude Code, run one command to add the Claude Imagine MCP server. In claude.ai or Desktop, add it as a custom connector by URL.
Use SKILL.md to define prompt structure, file naming, review steps, and when Claude should call generate_image.
Pass a model — nano-banana-2, gpt-image-2, seedream-4.5, flux-2-pro, or z-image — or let the default handle it. Credits come out of your shared balance.
Have another question? Email us at support@claudeimagine.com.
Runtime details verified July 10, 2026. Sources: Anthropic Agent Skills documentation and Claude custom connector documentation.
Free credits included — connect once and start generating.