MCP & Skill

Let agents generate images, videos and audio with WaveAPI. To run Codex or Claude Code on WaveAPI text models, see Codex and Claude Code. Three ways to connect. Pick one per client:

Option	For	Needs
Remote MCP	Clients that support remote MCP (Codex, Claude Code, Cursor, …)	API key
Agent Skill	Clients that support skills, or when you don’t want to configure MCP	API key, Node.js 18+
AI docs	Agents that only read the docs and write calling code	Nothing

MCP and the skill call the same endpoints. Prices, validation and billing are the same as calling POST /v1/tasks directly.

Set the API key

Create an API key in the console and store it in the WAVEAPI_API_KEY environment variable. Do not put the key in commands, config files or chats. macOS (zsh):

read -s "WAVEAPI_API_KEY?WaveAPI API Key: "; echo
launchctl setenv WAVEAPI_API_KEY "$WAVEAPI_API_KEY"

launchctl setenv makes the variable visible to GUI apps such as Codex and Cursor; restart the app afterwards. For terminal use, also export WAVEAPI_API_KEY in ~/.zshrc.

Remote MCP

Item	Value
URL	`https://api.qingbo.ai/v1/mcp`
Transport	Streamable HTTP
Auth	`Authorization: Bearer <API key>`

codex mcp add waveapi \
  --url https://api.qingbo.ai/v1/mcp \
  --bearer-token-env-var WAVEAPI_API_KEY

claude mcp add --transport http waveapi https://api.qingbo.ai/v1/mcp \
  --header 'Authorization: Bearer ${WAVEAPI_API_KEY}'

{
  "mcpServers": {
    "waveapi": {
      "url": "https://api.qingbo.ai/v1/mcp",
      "headers": { "Authorization": "Bearer ${env:WAVEAPI_API_KEY}" }
    }
  }
}

The Claude Code command uses single quotes, so ${WAVEAPI_API_KEY} is stored as written and read from the environment at run time. Test it with:

Use WaveAPI to list 5 video models. Don't generate anything.

Tools

Tool	Does	Billed
`list_models`	Lists the image, video and audio models this key can use; filter by modality or keyword	No
`get_model_docs`	Returns the model’s documentation page as Markdown	No
`get_model_schema`	Returns the model’s capabilities (actions, parameters, allowed values, defaults, constraints) and list prices	No
`estimate_cost`	Prices a request without creating a task	No
`create_task`	Submits a generation task; same parameters as `POST /v1/tasks`	Yes
`get_task`	Returns task status and results	No
`cancel_task`	Cancels a queued task and refunds its reservation	No

create_task accepts idempotency_key. Retrying with the same key and the same input does not create a second task.
create_task accepts wait_seconds (0–45). A task that finishes within that time returns its result directly.
get_task returns terminal, should_poll and next_poll_after_seconds. Result URLs expire at expires_at (Unix seconds) in result.
The cost fields cost and pre_consumed_cost are in quota. cost_usd and pre_consumed_cost_usd give the USD amounts.

Agent Skill

npx skills add QingBo-AI/skills --skill waveapi-media

The skill ships a zero-dependency script that reads WAVEAPI_API_KEY; the client needs no MCP setup. See waveapi-media.

AI docs

Documentation index: https://docs.qingbo.dev/llms.txt
Append .md to any page URL for its Markdown source, e.g. https://docs.qingbo.dev/en/api-reference/video/seedance.md
For chat apps without web access, paste PROMPT.md at the start of the chat

Limits

Item	Rule
Models	Image, video and audio models called through `/v1/tasks`. For text models use the OpenAI-compatible endpoints
Media inputs	Public `http(s)` URLs, or `asset://` on models that support it. Local paths, `file://` and base64 are not accepted
Model scope	Follows the API key’s model restrictions

Developer Tools

Agent Tools

Set the API key

Remote MCP

Tools

Agent Skill

AI docs

Limits

​Set the API key

​Remote MCP

​Tools

​Agent Skill

​AI docs

​Limits

Set the API key

Remote MCP

Tools

Agent Skill

AI docs

Limits