AI/ML API MCP

AI/ML API MCP is a remote MCP server from AI/ML API (AIMLAPI) at mcp.aimlapi.com that connects an agent to the company's catalog of models and tools. The vendor describes the catalog as 10,000+ models and tools, covering chat, image, video, audio and embedding models plus tools for web search, page extraction, screenshots, social media, news and company data.

What It Is

The product is a single MCP connector that lets an agent search, compare and run models and tools from within a chat. Users sign in once with an AI/ML API account through the browser, so no separate provider keys are needed. An API key passed as a Bearer header is supported as an alternative.

How the agent picks models and tools

For tools, the agent calls tools_search to find candidates, tools_inspect to see a tool's input and price, then tools_run to execute it. For models, it can search and compare the catalog by price and context window, for example asking for the cheapest model with function calling and 128k context. Catalog searches, input and price checks, and run-status checks are described as free, and each tool run is capped by a max_cost_usd setting.

Setup path

Claude (Desktop and claude.ai), Codex, Cursor and Claude Code have step-by-step instructions that add the MCP URL as a custom connector or server and then sign in. Other clients work if they support remote MCP over Streamable HTTP with OAuth 2.1 + PKCE. Hermes Agent, Codex, OpenCode, Cline and other coding agents can also connect through the aimlapi CLI, for example with npx aimlapi hermes.

Safety and error handling

Tool results come from the open web and are returned marked untrusted, so agents treat them as data rather than instructions. Failed provider calls return is_error: true, retryable errors can be retried, and a repeated call with the same idempotency_key and input returns the same run without charging again.