n8n Agents: give an agent a Sume video workflow
n8n's new Agents can use workflows and MCP servers as tools. A workflow that calls Sume keeps the API key out of the agent's hands.
In n8n Agents, give the agent a workflow as a tool: a small workflow that calls Sume's POST /v1/videos with a fixed model and a bounded prompt, and mark that tool for approval. n8n's announcement says each tool runs with the credential attached to it and that the agent never holds the keys, so your Sume API key stays inside the workflow.
The n8n facts come from its September 25, 2026 announcement, read 2026-09-29; it describes agents as still in preview. The Sume facts come from Video Generation and MCP tools and gates. Sume has no n8n connector; the workflow uses a plain HTTP Request node.
What can an n8n Agent use as a tool?
The announcement lists tools as built-in n8n integrations, MCP servers, and any workflow. For a paid video API two of those apply.
| Tool type | What n8n says | For a Sume video |
|---|---|---|
| Workflow | A whole process you've already built, run exactly the way you defined it | A workflow with one HTTP Request node to POST /v1/videos |
| MCP server | The agent gets all of its tools at once, and you can exclude any you don't want it to touch | Sume's hosted MCP at https://mcp.sume.com/mcp; exclude tools you do not want |
Why put the Sume call in a workflow?
The announcement's own example is a workflow that logs a note: without it, the agent would need write access to the CRM and instructions to keep to one field, and with it "the agent never holds that credential". The same shape fits a paid API. The workflow fixes what leaves your system: the model such as sume/auto and a length limit, so the agent cannot pick them. Generate a fresh Idempotency-Key per run inside the workflow.
Sume bills a video at provider list times 1.25, reserved on submit, so the fixed parameters also bound what each call can cost.
How do I keep a person in the loop?
The announcement says you can mark a tool as sensitive, and the agent then pauses for Approve or Reject before using it. Mark the video workflow that way. Its own advice for anything sensitive is to start where the blast radius is small: scoped tools, a test channel, and approvals on any action that writes to a system of record.
Approval and idempotency solve different problems. Sume's Idempotency-Key makes a retried submit return the original job; the docs describe the MCP idempotency_key as "for transport/dedup, not human approval". You want both.
Can I use Sume's hosted MCP instead?
Yes, as the announcement's MCP server option. A Sume API key gives the full hosted tool set, which includes paid tools, so the exclusion list matters. Sume's MCP docs say paid tools require an idempotency_key, and max_spend_usd is optional and enforced only when provided. The workflow route instead sets those values in a node you control.
What does it cost to run the agent?
The announcement says one agent turn is one execution, tool calls to your workflows do not count separately, and agents share your workflow execution quota. That covers n8n's side only. Each Sume job is billed by Sume; see how much does an AI video cost for the math.
Because agents are in preview, test the workflow tool with approvals on before you publish the agent.
Sources
Related posts
More in Integrations
- n8n AI agent HTTP Request tool for video jobs
Attach the HTTP Request node to an n8n AI agent as a tool and use Optimize Response to hand the model only a job's status and result URL.
- n8n error workflow: make it fire when an API job fails
An n8n error workflow runs only when an execution fails. A remote job that ends failed is a normal 200 read, so check its status and throw with Stop And Error.
- n8n HTTP Request timeout: what to do when a job takes minutes
The n8n HTTP Request node's Timeout option aborts slow responses. For jobs that take minutes, submit async, then poll with a Wait node.
- n8n Loop Over Items: batch paid API calls under a rate limit
n8n's Loop Over Items node sends items through in batches of Batch Size, then emits all results on done. Pair it with Wait to pace a paid API.
Written by Sume