MCP media tools

Add one URL to Claude Code, Cursor, VS Code, Codex, Claude, ChatGPT or another MCP client, sign in in the browser, and generate images, video, music and speech, and check your balance and call log

https://qd.zsopc.com/v1/mcp is an MCP server (Streamable HTTP) that turns the platform's image, video, music and speech APIs into 9 tools. Add it to an MCP client such as Claude Code, Cursor, VS Code, Codex, Claude or ChatGPT and use them in plain language.

No API key to copy: add the URL in your client, the client opens the browser, you sign in to your 小战AIhub account and click Allow. After that it renews itself. Billing, rate limits and the call log are the same as calling the API directly; usage is billed to the signed-in account.

Knowledge-base search is a separate MCP server (https://qd.zsopc.com/mcp) with the same sign-in; see Knowledge base API & MCP. You can add both.

Tools#

ToolWhat it doesCost
list_modelsModels you can use, their category and price (credits; ¥1 = 60 credits), filterable by category or keywordFree
generate_imageText to image, or edit from reference images; waits up to 25 s and returns image links when done, otherwise a task idPer image
create_videoText to video, or from a first frame, first and last frames, or reference images; returns a task id right awayBy model, resolution and length
generate_musicTwo songs (with lyrics and cover art) from a description or your lyrics; returns a task id right awayPer request
text_to_speechText to speech (MP3) with a preset voice or one you cloned; waits up to 25 sPer character
get_taskStatus and result of an image, video, music or speech taskFree
list_voicesPreset voices and the voices you clonedFree
get_balanceCredits leftFree
search_logsThe call log: model, status, cost, errors, task resultsFree
  • Images and video need a model (the model calls list_models first to compare prices); music defaults to suno-v6, speech to voice-tts-pro.
  • Reference images and first / last frames must be public http(s) links; put a local image somewhere publicly reachable first.
  • Failed tasks are refunded automatically.

Setup#

Every client below needs only the URL https://qd.zsopc.com/v1/mcp; the first time you use it, sign in in the browser when asked. The name xiaozhanai is up to you.

Claude Code#

claude mcp add --transport http --scope user xiaozhanai https://qd.zsopc.com/v1/mcp

--scope user makes it available in every project (without it, only in the current one). Then type /mcp in Claude Code, pick xiaozhanai (it says it needs authentication) and choose Authenticate; the browser opens the sign-in page. Or run claude mcp login xiaozhanai in a terminal.

Cursor#

Global ~/.cursor/mcp.json, or .cursor/mcp.json in a project:

{
  "mcpServers": {
    "xiaozhanai": {
      "url": "https://qd.zsopc.com/v1/mcp"
    }
  }
}

Find it under Customize → MCPs in the sidebar and follow the sign-in prompt. In the Cursor CLI: agent mcp login xiaozhanai.

VS Code (GitHub Copilot)#

For all workspaces use ~/.copilot/mcp-config.json; for one workspace, .mcp.json at the workspace root:

{
  "mcpServers": {
    "xiaozhanai": {
      "type": "http",
      "url": "https://qd.zsopc.com/v1/mcp",
      "tools": ["*"]
    }
  }
}

The browser opens for sign-in on the first connection; if it doesn't, open the server's menu under MCP Servers and choose Authenticate. Needs VS Code 1.101 or later. If you still use .vscode/mcp.json, write {"servers": {"xiaozhanai": {"type": "http", "url": "https://qd.zsopc.com/v1/mcp"}}} (deprecated since VS Code 1.146).

Codex#

codex mcp add xiaozhanai --url https://qd.zsopc.com/v1/mcp
codex mcp login xiaozhanai

Or write ~/.codex/config.toml yourself and run codex mcp login xiaozhanai:

[mcp_servers.xiaozhanai]
url = "https://qd.zsopc.com/v1/mcp"

The ChatGPT desktop app and the IDE extension read the same file; you can also add it under their Settings → MCP servers and click Authenticate. On a remote machine use codex mcp login xiaozhanai --no-browser.

Claude (web, desktop, mobile)#

Customize → Connectors → Add → Add custom connector, enter a name and the URL https://qd.zsopc.com/v1/mcp, choose the sign-in options when asked (Use Claude's published identity is recommended), add it and click Connect, then sign in on the page that opens. Free accounts can add one custom connector. On Team / Enterprise an organization admin adds it under Organization settings → Connectors and members Connect themselves. Connectors added on the web or desktop also work on mobile.

ChatGPT#

At https://chatgpt.com/plugins, click the plus button, choose Add custom MCP server, enter a name and the Server URL https://qd.zsopc.com/v1/mcp, choose OAuth for authentication, create and install it, then type @ in a chat to pick it. Which plans can add custom MCP servers is up to ChatGPT; see its help center (currently mainly Business, Enterprise and Edu workspaces).

Gemini CLI#

~/.gemini/settings.json:

{
  "mcpServers": {
    "xiaozhanai": {
      "httpUrl": "https://qd.zsopc.com/v1/mcp"
    }
  }
}

Then run /mcp auth xiaozhanai in Gemini CLI.

Other clients#

A client that supports remote (Streamable HTTP) MCP servers and OAuth sign-in needs only the URL https://qd.zsopc.com/v1/mcp; the sign-in follows the standard MCP authorization flow.

These setups follow each client's official documentation as of November 2026; if a client changes, its documentation wins.

With an API key#

Where there is no browser (CI, servers), with a client that cannot sign in through OAuth, or when connecting through the China entry zsopc.com, use an API key instead: create one under API Keys and send it as Authorization: Bearer sk-gpushare-xxx (or x-api-key: sk-gpushare-xxx). Keys in the URL are not accepted.

claude mcp add --transport http xiaozhanai https://qd.zsopc.com/v1/mcp \
  --header "Authorization: Bearer sk-gpushare-xxx"
{
  "mcpServers": {
    "xiaozhanai": {
      "url": "https://qd.zsopc.com/v1/mcp",
      "headers": { "Authorization": "Bearer ${env:XIAOZHANAI_API_KEY}" }
    }
  }
}

In Codex use bearer_token_env_var = "XIAOZHANAI_API_KEY" (the name of the environment variable), or codex mcp add xiaozhanai --url https://qd.zsopc.com/v1/mcp --bearer-token-env-var XIAOZHANAI_API_KEY.

Using it#

Just ask: "Draw a poster with gpt-image-2 about …", "Make a 5-second 720p video of …, and give me the link when it's done". The model picks a model with list_models, then calls the generation tool; video and music take a few minutes, so the model checks back with get_task.

FAQ#

  • How long does the sign-in last? The connection renews itself and stays valid while you use it; you only sign in again after 90 days without use (and if the browser is still signed in, that's one more click on Allow).
  • How do I disconnect? Under My connections in the console you see every app signed in through the browser, when it was last used and what it spent this month; Disconnect takes effect at once. Resetting your password disconnects all of them, and so does deleting the account.
  • Can enterprise employees use it? Yes, with the same rule as Deski desktop: when the enterprise administrator turns off an employee's Deski access in the enterprise console, that employee's connections are disconnected too, and new ones are refused.
  • The sign-in page says "Unverified app"? The app is not on the recognized list. Make sure you just started this sign-in from your own client and check the return address shown on the page before clicking Allow.

Notes#

  • Billed like a direct call: each generation is charged at the model's price to the signed-in account (or the account of the API key). The tools' idempotency_key argument means the same as the API's Idempotency-Key header: the same key with the same arguments is not charged twice.
  • Rate limits: the same concurrency and per-minute limits as an ordinary API key.
  • When the balance runs out: get_balance, list_models and search_logs keep working; checking a task (get_task, and the wait inside the generation tools) and listing voices (list_voices) need a balance above zero, like the direct API calls.
  • Not tools yet: digital humans, voice cloning and smart editing; call the media APIs directly.