Vyrl

Point Claude Code, Codex, Cursor, or another bearer-token MCP client at Vyrl. Build canvases, run Flows, and pull back finished video without leaving the conversation.

Connect MCP and start creatingFree

1

Copy the Vyrl MCP URL

The same endpoint for every client. Streamable HTTP, no OAuth flow.

https://vyrl.pro/api/mcp

Add it in.mcp.json

Reference a token from your environment so the credential never lands in shell history or process arguments. Set VYRL_MCP_TOKEN through your secret manager before starting Claude Code.

{
  "mcpServers": {
    "vyrl": {
      "type": "http",
      "url": "https://vyrl.pro/api/mcp",
      "headers": {
        "Authorization": "Bearer ${VYRL_MCP_TOKEN}"
      }
    }
  }
}

3

Create a token and start

Mint a scoped token, drop it into the header, then ask your agent to open a canvas or run a Flow.

Create MCP token

Example MCP client configurations for the preconfigured bearer-token endpoint. MCP

Copy a prompt, paste it into your agent.

Each one uses tools this server really exposes. Connect first, then paste.

vyrl · agent session

you

Explore what you have

“List my Vyrl canvases and summarise what each one is set up to do.”

list_canvases get_canvas

Build a node

“In my Launch Reel canvas, add a Seedance video node, wire it to the prompt node, and show me the graph before running anything.”

get_canvas_node_catalog update_canvas

Run and wait

“Run my Weekly Recap flow, then poll until it finishes and give me the output link.”

list_flows run_flow get_flow_run

Start from scratch

“Create a new Vyrl canvas called Product Teaser with an image node feeding a video node.”

create_canvas update_canvas

Run your configured models from your agent.

A Flow run over MCP executes the model configuration already saved in your Canvas. Vyrl handles the provider, the credits, and the durable result.

• sora-2-pro
• flux-2-pro
• nano-banana-pro
• ray-2
• imagen-4-ultra
• ideogram-v3-quality
• qwen-image
• recraft-v4.1-pro

A token only does what you granted.

Reading, running, and editing are separate. An agent wired for research cannot start a paid run, and cannot touch your work.

canvas:read 6 tools

Look, never touch

Inspect canvases, Flows, runs, and the node catalog. A read-only token cannot change or start anything.

canvas:run 1 tool

Start work that costs credits

Run a Flow and read its durable result. Runs draw on your credit balance, which is why this is its own grant.

canvas:edit 3 tools

Change what you built

Create, update, and delete canvases. Deleting is opt-in - a new token does not get it by accepting defaults.

10 tools, and the scope each one needs.

• list_canvases canvas:read
• get_canvas canvas:read
• get_canvas_node_catalog canvas:read
• create_canvas canvas:edit
• update_canvas canvas:edit
• delete_canvas opt-in canvas:edit
• list_flows canvas:read
• get_flow canvas:read
• run_flow canvas:run
• get_flow_run canvas:read

Irreversible tools are excluded from a new token's defaults and must be selected deliberately.

Before you connect.

What is MCP?

Model Context Protocol is an open standard that lets an AI agent call external tools. Connecting Vyrl means your agent operates your Canvas directly, instead of you shuttling things between a chat window and the app.

Do I need an API key?

You need a Vyrl MCP token from your profile. It is a bearer token rather than an OAuth app, so the Authenticate button in most MCP client UIs does not apply here.

Can an agent delete my work?

Only if you let it. Deleting requires canvas:edit, and delete_canvas is excluded from a new token's defaults, so it has to be chosen deliberately.

Does running a Flow cost credits?

Yes. A run started over MCP draws on the same credit balance as one started in the app, which is why canvas:run is separated from canvas:read.