MCPremote MCP server

Generate AI images and video in Claude

Connect APIPod MCP to Claude, ChatGPT, Cursor, Grok or WorkBuddy and generate and edit AI images and video directly inside your agent workflow.

any MCP client — no SDK, no proxy

OAuth 2.1 · PKCE S256 · scopes profile email · 7 toolsplugin source on GitHub (Apache-2.0)

A cobalt-blue ceramic mug on a pale gray studio backdrop, soft window light, product photography

30-second cinematic aerial action spectacle: a battleship cuts through storm waves as a fighter jet bursts from volumetric clouds

01/clients

pick a client for its exact config

Connect your client.

Every client reads the same two discovery documents: RFC 9728 at /.well-known/oauth-protected-resource/mcp, RFC 8414 at /.well-known/oauth-authorization-server. Hosted clients redirect to an allowlisted domain; CLIs call back on localhost, so no port has to be registered.

Anything else with an MCP client and OAuth support uses the same URL.

Claude

hosted redirect

Hosted redirect to claude.ai/api/mcp/auth_callback. Nothing to install.

01
Settings → Connectors → Add custom connector.
02
Paste the URL and continue; Claude reads the auth metadata on its own.
03
Sign in with your APIPod account.

url

https://api.apipod.ai/mcp

redirect: claude.ai/api/mcp/auth_callback
endpoint: api.apipod.ai/mcp
auth: OAuth 2.1 · PKCE S256
scopes: profile · email

02/tools

GET /mcp/catalog →

Seven tools, two of which spend money.

list_models, get_model_params, estimate_cost, get_balance and check_job are free. generate_image and generate_video reserve balance, and are refunded when the provider call fails.

list_modelsimage · video · speechfree

get_model_paramsrequest schema, enums, defaultsfree

estimate_costUSD price vs. balancefree

get_balanceremaining USDfree

generate_imagejob_idpaid

generate_videojob_idpaid

check_jobstatus, permanent URLs, final costfree

Generation tools return a job_id. check_job polls it and returns permanent URLs plus the final cost in USD.

Any other MCP client

If it speaks Streamable HTTP and OAuth, it works — add the server by URL:

{
  "mcpServers": {
    "apipod": {
      "type": "http",
      "url": "https://api.apipod.ai/mcp"
    }
  }
}

The server publishes its own auth metadata, so there is no client ID to paste.

03/models

All 128 models →

Models you can call today.

The agent picks from the same catalog as the REST API. Each card links to that model's page with its full spec and per-second or per-request price.

Image

Wan 2.7 Image Spicy$0.03 /call Qwen Image Spicy$0.015 /call Seedream 5.0 Flash$0.018 /call Nano Banana 2.1$0.02 /call Seedream 5.0 Pro$0.035 /call GPT Image 2.5$0.02 /call

+4 more in this category

Video

MiniMax H3 Lite$0.04 /sec Kling 4.0$0.085 /sec Kling 3.0$0.08 /sec Wan 3.0 Video Prime Spicy$0.10 /sec Wan 3.0 Video Pro Spicy$0.20 /sec Wan 3.0 Video Spicy$0.06 /sec

+24 more in this category

04/billing

How a generation is billed.

Prices come from the same table as the REST API, in USD. No credits, no subscription for the connection.

quote first

estimate_cost returns a USD price for the exact request and whether the balance covers it.

one call, one charge

Each generate_* call reserves its quote. Re-running, comparing models or retrying a bad result is another charge, and the skills ask before spending.

refunded on failure

A provider-side failure returns the reservation to your balance automatically.

per-client grants

Each client registers its own OAuth client, so one connection can be revoked in the console without touching the rest.

05/FAQ

Details that come up

The questions that decide whether this fits your setup.

Which clients are supported?

Claude, Claude Code, ChatGPT, Codex, Cursor, Gemini CLI, Grok, Muse Code, OpenClaw and WorkBuddy — each with the exact config on this page. Anything else that speaks Streamable HTTP and OAuth works with the same URL.

Which account does the client charge?

The APIPod account you sign in with during OAuth. Generations land in your request log with path=/mcp, like any REST call.

Do I need an API key?

No. Authentication is OAuth 2.1 with PKCE. The server provisions a key internally for logging and billing; nothing is pasted into the client.

Which models are available?

Every image and video model in the catalog — Seedance, Kling, Seedream, Nano Banana, GPT Image, Wan and the rest. list_models returns what your account can call right now.

Is speech supported?

Speech models appear in list_models, but generate_speech is not shipped yet — its output has to be persisted the same way images and video are. Use the REST API for TTS today.

Can I use my own images as references?

Yes, on models whose get_model_params lists image input: pass public HTTPS URLs in image_urls — the first is the primary (edit source or first video frame). Base64 is not accepted over MCP.

What if my image is only on disk?

Upload it in the APIPod console and use the URL it returns. Over MCP the server takes URLs only; there is no file upload tool.

How long does a generation take?

Images typically land in 10–40 seconds, video in 1–5 minutes. The client polls check_job for status; a finished job returns permanent URLs, so the result does not expire.

What can the client do?

Read the catalog and balance, and submit generations. Tokens carry profile and email scopes, PKCE is enforced, and each client's grant is revocable on its own.

Is there a rate limit?

Rate limits match your plan's REST API limits. There is no subscription for the MCP connection itself.

Generate AI images and video in Claude

Claude

Create an account, then connect the client.