AI Content Drop in Kimi Code

OAuth sign-in from consumer hosts is still being verified. Public browsing and quotes work without an account; clients that accept API-key headers can use an existing key. The OAuth steps below describe the connection flow to test once sign-in is ready. Registration alone does not prove a working sign-in.

Kimi Code CLI reads remote MCP servers from an mcp.json file and signs in to them with a slash command. Everything below is taken from the Kimi Code documentation for MCP servers and plugins; where that documentation is silent, so is this page.

Add the server

Kimi Code reads mcp.json at two levels:

user level: ~/.kimi-code/mcp.json (or $KIMI_CODE_HOME/mcp.json)
project level: .kimi-code/mcp.json in the working directory

A project entry overrides a user entry with the same name. Add:


{
  "mcpServers": {
    "aicontentdrop": {
      "url": "https://aicontentdrop.com/mcp"
    }
  }
}

A url field means the HTTP transport, which is what our server speaks. Do not add "transport": "sse"; that selects the legacy transport.

You can also add or edit the entry interactively with /mcp-config inside the CLI, and /mcp shows the connection status of every configured server.

Sign in with OAuth

The four public tools (list_models, estimate_credit_cost, search_articles, get_article) work as soon as the server connects. For the account tools, run:


/mcp-config login aicontentdrop

Kimi Code opens a browser for the sign-in. Approve the consent screen on aicontentdrop.com and the account tools start working. No key is pasted anywhere.

Developer fallback: an API key header

If you would rather not use OAuth, for example in a script or on a machine with no browser, create a key at Settings → Integrations and send it as a header. Kimi Code supports a static headers map, or a bearerTokenEnvVar that names an environment variable holding the token:


{
  "mcpServers": {
    "aicontentdrop": {
      "url": "https://aicontentdrop.com/mcp",
      "bearerTokenEnvVar": "ACD_API_KEY"
    }
  }
}


export ACD_API_KEY=acd_live_YOUR_KEY

The environment-variable form keeps the key out of the config file. The literal form is "headers": { "Authorization": "Bearer acd_live_YOUR_KEY" }; treat the file as a secret if you use it.

First generation

> Quote a 5-second Kling 3.0 clip at 1080p, then generate one of a matcha tin > on a linen backdrop, and poll it until it is done.

The agent calls estimate_credit_cost (which returns a quote_id valid for 15 minutes), then generate_video with that quote_id, then get_generation until the job reports completed. Credits are charged on success only, and a retry with the same quote_id returns the original job rather than a second charge.

Bundle it as a plugin

Kimi Code plugins can declare MCP servers in their manifest, so a team can install the connection with one command instead of editing mcp.json. The manifest lives at <plugin_root>/kimi.plugin.json (or <plugin_root>/.kimi-plugin/plugin.json); name is the only required field:


{
  "name": "aicontentdrop",
  "description": "AI video and image generation on one credit balance",
  "mcpServers": {
    "aicontentdrop": {
      "url": "https://aicontentdrop.com/mcp"
    }
  }
}

Install it from a local directory, a ZIP URL, or a GitHub URL:


/plugins install ./aicontentdrop-plugin
/plugins install https://github.com/<owner>/<repo>

Plugin changes apply after /reload or in a new session. Bundled servers can be switched with /plugins mcp enable <id> <server> and /plugins mcp disable <id> <server>; the plugin itself with /plugins enable <id>, /plugins disable <id> and /plugins remove <id>. Installed plugins live under $KIMI_CODE_HOME/plugins/managed/<id>/.

Troubleshooting

/mcp shows the server but no tools. Check the URL is https://aicontentdrop.com/mcp with no trailing slash and that the entry has no "transport": "sse".

Account tools answer AUTH_REQUIRED. Run /mcp-config login aicontentdrop again, or check the header if you chose the API-key route. The public tools never need either.

QUOTE_REQUIRED, QUOTE_EXPIRED or QUOTE_MISMATCH. Every generation needs a fresh quote_id from estimate_credit_cost for the same model, duration and resolution. Quote again and retry.