Integration

Volcengine Ark integration

Volcengine Ark (Huoshan Fangzhou, ByteDance's model platform for mainland China) hosts Doubao, DeepSeek, GLM and Kimi models. An agent can chat with a model (text, image, video, audio and PDF input, function calling, deep thinking, JSON output), run stateful Responses API conversations and read or delete stored responses, upload, list, read and delete input files, generate images with Seedream, create video tasks with Seedance and image-to-3D tasks with Seed3D and then read, list, cancel or delete them, create multimodal embeddings and tokenize text. Every model call is billed by Volcengine (per token, per generated image, per video or 3D token), and a model must be activated in the Ark console before use. Not included: streaming output, the control-plane APIs (endpoints, API keys, batch jobs, fine-tuning, signed with Volcengine access keys), the Batch Chat API (marked for removal), the beta System One API, the Anthropic-compatible Messages API, Managed Agents, and file download links (temporary credentials). Authentication: an Ark API key from the Ark console's API Key page, sent as Authorization: Bearer.

Connect Volcengine Ark How integrations work

Ask it in chat

Say what should happen.
The workflow builds itself.

When this happens, do that in Volcengine Ark.Built from the 17 Volcengine Ark tools below, on the connection you add once.

What you can do with Volcengine Ark

Each of these is one operation on one endpoint, available both as a workflow step and as a tool an AI agent can call on its own. Nothing bundles several operations into one action, so what a run did stays readable afterwards.

Each badge shows the default agent permission of an endpoint (allowed: an agent calls it without asking; ask first: you are asked before an agent calls it, when the agent or chat asks permission for sensitive actions; never: no agent may call it), which every user can change endpoint by endpoint in their credential settings.

Most endpoints run on the free plan. A few, mainly the ones that publish on your behalf, need a paid plan.

File, 4 operations

delete_fileDelete fileDELETEAgents: ask firstDELETES a file for good (no restore): it can no longer be used as model input. Returns deleted true when removed.
get_fileGet fileGETAgents: ask firstReturns a file's metadata: status (processing, active or failed, with error), name, size, MIME type, purpose and expiry. Call it after upload_file until status is active. Download links are not returned.
list_filesList filesGETAgents: ask firstLists your uploaded files, 100 per page by default, optionally by purpose; page with after set to last_id. Download links are not returned.
upload_fileUpload filePOSTAgents: ask firstUploads a file, or registers a public URL, as model input and returns its id with status processing; poll get_file until active before using it. Kept 7 days by default (1 to 30). Storage in your own TOS bucket and video preprocessing are not offered.

Generation task, 3 operations

delete_generation_taskDelete generation taskDELETEAgents: ask firstCancels a queued generation task, or DELETES the record of a succeeded, failed, cancelled or expired one (it can no longer be read). A running task cannot be cancelled. Answers with no body.
get_generation_taskGet generation taskGETAgents: ask firstReturns a video or 3D generation task: status (queued, running, succeeded, failed, cancelled, expired), result URLs (valid 24 hours), settings, error and token usage. Call it again until the status is final.
list_generation_tasksList generation tasksGETAgents: ask firstLists video and 3D generation tasks, 20 per page by default, filtered by status, task ids, inference endpoint or service tier. Returns items and total.

Response, 3 operations

create_responseCreate responsePOSTAgents: ask firstGenerates a model response with the Responses API (billed per token). Stored for 3 days by default, so previous_response_id can continue the conversation. Supports deep thinking, functions, web search and knowledge-base tools.
delete_responseDelete responseDELETEAgents: ask firstDELETES a stored response: it can no longer be read or continued with previous_response_id. Returns deleted true or false.
get_responseGet responseGETAgents: ask firstReturns a stored response by id: status, output items, usage and the settings used. A response still being generated answers an error.

3d generation task, 1 operation

create_3d_generation_taskCreate 3d generation taskPOSTAgents: ask firstStarts a billed Seed3D image-to-3D task from one image and returns its id; poll get_generation_task until status is succeeded (content.file_url, kept 24 hours), failed, cancelled or expired.

Chat completion, 1 operation

create_chat_completionCreate chat completionPOSTAgents: ask firstGenerates the next assistant message of a conversation with a model hosted on Ark (Doubao, DeepSeek, GLM, Kimi). Billed per token by Volcengine. Takes image, video, audio and PDF input; supports function calling, deep thinking and JSON output.

Embedding, 1 operation

create_multimodal_embeddingsCreate multimodal embeddingsPOSTAgents: ask firstTurns texts, images or videos into vectors for search and matching (billed per token). Returns the dense embedding, plus a sparse vector or token-level vectors when enabled.

Image, 1 operation

generate_imagesGenerate imagesPOSTAgents: ask firstGenerates images with Seedream (billed per generated image) and returns their URLs (valid 24 hours), or splits one image into layers. With sequential_image_generation=auto a group of up to 15 images may be generated and billed.

Response input item, 1 operation

list_response_input_itemsList response input itemsGETAgents: ask firstLists the input items of a stored response (the context it was generated from), 20 per page by default; page with after set to last_id.

Text, 1 operation

tokenize_textTokenize textPOSTAgents: ask firstSplits one or more texts into tokens with a model's tokenizer: token ids, character offsets and the token count of each text. Text only.

Video generation task, 1 operation

create_video_generation_taskCreate video generation taskPOSTAgents: ask firstStarts a billed Seedance video generation task and returns its id; poll get_generation_task until status is succeeded (content.video_url), failed, cancelled or expired. Billed in video tokens; duration drives the cost.

Use Volcengine Ark from Claude Code, Cursor or Codex

Connect Volcengine Ark in LiveContext, create an API key in Settings, then add the LiveContext MCP server to your client. Its Volcengine Ark operations become tools the client can call, next to every other integration you connected.

Claude Code

claude mcp add --transport http livecontext https://livecontext.ai/mcp --header "X-API-Key: YOUR_API_KEY"

Codex (~/.codex/config.toml)

[mcp_servers.livecontext]
url = "https://livecontext.ai/mcp"

[mcp_servers.livecontext.http_headers]
"X-API-Key" = "YOUR_API_KEY"

Cursor, Claude Desktop and other JSON configs

{
  "mcpServers": {
    "livecontext": {
      "type": "http",
      "url": "https://livecontext.ai/mcp",
      "headers": {
        "X-API-Key": "YOUR_API_KEY"
      }
    }
  }
}

Set up the MCP server step by step

How to connect Volcengine Ark

Create your workspace
Sign up free. Nothing to install.
Connect Volcengine Ark once
Add your Volcengine Ark connection in Settings. Every tool from this integration reuses it, and secrets are encrypted at rest and visible only to you.
Describe the job in chat
The workflow builds itself in front of you and picks the endpoints it needs. Drag the steps yourself on the canvas whenever you would rather be explicit.