Add every model to Claude Code with the ApexApi MCP

The Model Context Protocol lets an agent or editor call external tools. The ApexApi MCP turns the whole platform into those tools: your agent can talk to any model, generate an image or a video, synthesize speech, or read a web page, without you writing a single integration. Wire it once and it is there in every session.

Here is how to connect it.

Connect it with OAuth (easiest)

If you use Claude Code, one command adds it:

claude mcp add --transport http apexapi https://api.apexapi.dev/mcp

Then run /mcp in Claude Code, choose apexapi, and click Authenticate. A browser opens to the ApexApi consent screen; approve it, and the tools are live. The token you get back is a dashboard-revocable key, so you can cut access any time from apexapi.dev/keys.

On claude.ai: Settings, then Connectors, then Add custom connector, and enter the URL https://api.apexapi.dev/mcp, then Connect and approve.

Or run it locally over stdio

For clients that take a local command (or if you prefer stdio), run the apexapi-mcp package over npx. Get a key at apexapi.dev/keys, then add this to your MCP client config:

{
  "mcpServers": {
    "apexapi": {
      "command": "npx",
      "args": ["-y", "apexapi-mcp"],
      "env": { "APEXAPI_API_KEY": "ak-..." }
    }
  }
}

There is also a bearer-key remote option for clients that pass a header directly:

claude mcp add --transport http apexapi https://api.apexapi.dev/mcp --header "Authorization: Bearer ak-..."

What your agent can now do

Once connected, these tools show up in your client:

• chat and list_models: talk to any model in the catalog, or list what is available.
• generate_image, generate_video, generate_speech: create media from a prompt.
• scrape_page, crawl_site, extract_structured: give the agent live web context.
• get_balance: check your remaining credits.

All of it runs on the same key and the same prepaid balance as the API.

Worked example: generate an image from your editor

Ask your agent something like: "Generate a product image of a brushed-metal API key card on a dark studio surface." It calls generate_image with openai/gpt-image-2 and hands back a hosted URL. This image was generated exactly that way, from an agent, through the MCP:

No leaving the editor, no separate image tool, no second key.

Prompt it well

The tools are only half of it. The other half is the prompt. If you are going to generate images from your agent, the same rules apply as anywhere else. Our GPT Image 2 prompting guide has five copy-paste techniques (including the text-rendering trick) that make a real difference.

Where to go next

• MCP docs: the full reference, tools, and options.
• Model catalog: everything the MCP can reach.
• GPT Image 2 prompting guide: get sharper images from your agent.

One connector, every model, right where you already work.