Visual overview

Use the workflow to follow the task, and the architecture map to separate responsibilities. These are conceptual maps; the guide below defines implementation details and verification limits.

01 / WorkflowFrom intent to a checked result
  1. Connect the local iOS Agent server
  2. Add the hosted AI/ML API connector
  3. Sign in with your own account
  4. Generate concepts, then build assets locally
02 / ArchitectureResponsibility boundaries
  1. Boundary 1Your MCP client
  2. Boundary 2iOS Agent MCP (local)
  3. Boundary 3AI/ML API MCP (hosted, user account)

Connected responsibilities, not a required class hierarchy or an execution trace.

AI/ML API publishes a hosted MCP server at https://mcp.aimlapi.com/mcp that exposes its model gateway: model discovery, chat, embeddings, and image, video and audio generation, billed to the user's own AI/ML API account. It can run in the same MCP client as iOS Agent MCP. Nothing in this repository calls it, bundles it, resells it or stores its credentials; the iOS Agent server works without it.

This page covers running the two servers side by side and the one iOS workflow where the pairing is useful. It is not an endorsement of a provider or a claim that any AI/ML API model produces App Store-ready assets.

Who it is for

Developers who already hold an AI/ML API subscription or balance and want generation capabilities the iOS Agent server does not provide, mainly app icon concepts, placeholder imagery and short sounds for prototypes. The account, sign-in and charges belong to the user. Without an AI/ML API balance the connector lists models but billable inference fails; that is the provider's behavior, not an iOS Agent error.

Connect both servers

Authentication is OAuth through the browser by default. Do not paste an AI/ML API key into any file in your app repository or into this project.

Claude Code:

claude mcp add ios-agent -- npx -y ios-agent-mcp@latest
claude mcp add --transport http aimlapi https://mcp.aimlapi.com/mcp

Run /mcp, select aimlapi and choose Authenticate. Add --scope user to the second command to make the connector available in every folder.

Claude Desktop or Claude web: add https://mcp.aimlapi.com/mcp as a custom connector under Settings → Connectors and click Connect. Keep the local iOS Agent stdio entry from installation.md; the two do not replace each other.

Cursor: merge aimlapi-cursor.json into ~/.cursor/mcp.json or .cursor/mcp.json. Leave out any headers block so OAuth is used, then enable and log in from the MCP settings pane.

Other clients need remote MCP over Streamable HTTP with OAuth 2.1, PKCE and dynamic client registration. If a client only supports bearer headers, the user supplies their own key through the client's secret storage, never through this repository.

When to use which server

Task Server
Swift review, Apple references, simulator, /ios-build, screenshots iOS Agent MCP
Generate an icon concept, placeholder image, mock audio AI/ML API (user's account)
Turn generated artwork into a layered .icon or an asset catalog iOS Agent MCP: icon-composer.md, asset-generation.md
Check a generated palette against contrast rules iOS Agent MCP: review_color_system

A typical flow: describe the app, let the agent plan it with /ios-build, ask the AI/ML API connector for a few icon concepts, pick one, then route the chosen file through the icon and asset guides above. Generated artwork is the user's to review for licensing, trademark and App Review suitability; neither server checks that.

Boundaries

  • iOS Agent MCP never sends app source, prompts, logs or paths to AI/ML API, and this project adds no provider routing for it. The ai ask helper is unchanged.
  • The connector's tools can spend the user's balance. Review tool permissions before allowing autonomous calls, the same as for file-writing iOS Agent tools.
  • Removing the connector from a client does not revoke its token; revoke the app in the AI/ML API account for a clean reinstall.
  • Rate limits, model availability and pricing are the provider's and change without notice. Check the provider's pages rather than this guide for current numbers.

Sources

Checked October 28, 2026:

Recorded verification

None. This page documents the provider's published setup and the intended pairing. Connecting the hosted server requires an interactive OAuth sign-in with a personal AI/ML API account, which has not been exercised from this repository's test environment, so no tool discovery count or model session is claimed. Treat the steps above as unverified until you run them yourself.