Integration

VLM Run integration

VLM Run turns images, documents, audio and video into structured JSON. Extract typed data with a ready-made domain (invoices, receipts, resumes, captions, transcriptions; list_domains) or your own JSON schema or skill, run Orion visual agents (execute_agent, create_chat_completion), upload and manage files, read past predictions and executions, fetch artifacts the agents produce, manage reusable skills and send feedback. Every extraction and agent run is billed in credits on your VLM Run account. Not exposed: the OpenAI-compatible model and file mirrors, health checks, the two-step presigned upload, chat message history (it needs a session token) and the evaluation and skill-optimisation jobs. Authentication is a VLM Run API key (dashboard, Settings, API keys) sent as Authorization: Bearer <key>.

Connect VLM Run How integrations work

Ask it in chat

Say what should happen.
The workflow builds itself.

When this happens, do that in VLM Run.Built from the 33 VLM Run tools below, on the connection you add once.

What you can do with VLM Run

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.

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

Prediction, 7 operations

extract_audioExtract audioPOSTTranscribes audio (url or file_id) into timed segments. Billed in credits. On timeout the error names the id: see get_prediction. Waits for the job and returns the finished result; gives up after 10 minutes.
extract_documentExtract documentPOSTExtracts JSON from a document (url or file_id) with a domain or config.skills. Billed in credits. On timeout the error names the id: see get_prediction. Waits for the job and returns the finished result; gives up after 10 minutes.
extract_imageExtract imagePOSTExtracts JSON from images with a domain (list_domains) or config.skills/json_schema. Billed in credits. On timeout the error names the id: see get_prediction. Waits for the job and returns the finished result; gives up after 10 minutes.
extract_videoExtract videoPOSTExtracts a transcription or summary from a video (url or file_id). Billed in credits. On timeout the error names the id: see get_prediction. Waits for the job and returns the finished result; gives up after 10 minutes.
get_predictionGet predictionGETReturns one prediction by id: status (pending, enqueued, running, completed, failed or paused), response (the result) and usage.
get_prediction_domainGet prediction domainGETReturns the domain, model, input type and status of one prediction.
list_predictionsList predictionsGETLists your past predictions (extract_* jobs) with status, domain, response and credit usage, with skip/limit pagination.

File, 5 operations

delete_fileDelete fileDELETEDeletes an uploaded file by id. Only on Pro and Enterprise plans.
get_fileGet fileGETGet a file by ID.
get_file_by_hashGet file by hashGETChecks whether a file with this MD5 hash was already uploaded and returns it, so you can reuse its id instead of uploading again.
list_filesList filesGETGet all files uploaded by the user with pagination.
upload_fileUpload filePOSTUploads a file (image, document, audio, video, skill zip) and returns its id for the file_id of the extract_* tools, execute_agent or create_skill.

Skill, 5 operations

create_skillCreate skillPOSTCreates a skill from exactly one source: a prompt (and optional json_schema), an uploaded skill zip (file_id, then name and description are required) or a chat session_id.
get_skillGet skillGETReturns one skill (yours or a public one) by id: name, version, description and storage URI.
get_skill_download_urlGet skill download urlGETReturns a presigned download_url for the zip of one skill, valid for expires seconds.
list_skillsList skillsGETLists your organization's skills and the public ones (id, name, version, description). Set grouped=true for only the latest version of each name.
lookup_skillLookup skillPOSTFinds one skill by id, or by name with an optional skill_version (latest otherwise). Read-only.

Agent, 3 operations

create_agentCreate agentPOSTCreates a named agent from a prompt and optional JSON schema or skills; returns its id, name and status. Run it later with execute_agent.
list_agentsList agentsGETLists the agents of your organization with their id, name, description, prompt and output schema.
lookup_agentLookup agentPOSTFinds one agent by id, or by name (such as accounting-invoice:latest); returns its prompt, schema and sample. Read-only.

Agent execution, 3 operations

execute_agentExecute agentPOSTRuns an Orion agent or skill on your inputs; result in response. Billed in credits. On timeout the error names the id: see get_agent_execution. Waits for the job and returns the finished result; gives up after 10 minutes.
get_agent_executionGet agent executionGETReturns one agent execution by id: status, response (the result), usage (credits and cost) and timestamps.
list_agent_executionsList agent executionsGETLists agent executions of your organization (newest first per the vendor's pagination), with status, response and credit usage.

Feedback, 2 operations

list_feedbackList feedbackGETLists the feedback recorded for one prediction, agent execution or chat; set type to match the id.
submit_feedbackSubmit feedbackPOSTRecords feedback on a prediction, agent execution or chat: a corrected response and/or notes. Give exactly one of request_id, agent_execution_id or chat_id.

Artifact, 1 operation

get_artifactGet artifactGETDownloads one file an agent produced (image, video, audio, document) as a file: give its object_id (such as img_a1b2c3) and the execution_id or session_id it came from.

Chat, 1 operation

list_chatsList chatsGETLists your chat threads (session_id, title, summary, message count, timestamps), most recent first per the vendor.

Chat completion, 1 operation

create_chat_completionCreate chat completionPOSTChats with an Orion visual agent (OpenAI-compatible): send messages with text, image_url, video_url or file_url parts; the answer is choices[0].message.content. Billed per tokens and tools.

Domain, 1 operation

list_domainsList domainsGETLists every domain identifier the extract_* tools accept, such as document.invoice or image.caption. Answers without a key too.

Domain schema, 1 operation

get_domain_schemaGet domain schemaPOSTReturns the JSON schema, a sample answer and the GraphQL statement of one domain, so you know what extract_* will return for it. Read-only.

Hub, 1 operation

get_hub_infoGet hub infoGETReturns the version of the VLM Run hub (the catalog of domains).

Model, 1 operation

list_modelsList modelsGETLists the prediction models and the domain each one serves.

Skill version, 1 operation

create_skill_versionCreate skill versionPOSTCreates a new version of a skill from a new zip (file_id from upload_file): the new version gets its own id; the old one is kept.

Use VLM Run from Claude Code, Cursor or Codex

Connect VLM Run in LiveContext, create an API key in Settings, then add the LiveContext MCP server to your client. Its VLM Run 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 VLM Run

Create your workspace
Sign up free. Nothing to install.
Connect VLM Run once
Add your VLM Run 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.