Try it in ~30 seconds, no key required

The keyless demo gives you about $10 of demo credit per network address, at each model's real price — every model and every tool unlocked, nothing counted. Enough to see a real render before you sign up. Paste this config, no Authorization header, no account.

{
  "mcpServers": {
    "livepeer-agent": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://agent.livepeer.org/api/mcp", "--header", "X-Livepeer Agent-Tool-Profile: lean"]
    }
  }
}

Install per client

Endpoint: https://agent.livepeer.org/api/mcp

{
  "mcpServers": {
    "livepeer-agent": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://agent.livepeer.org/api/mcp",
        "--header", "Authorization:Bearer sk_YOUR_KEY",
        "--header", "X-Livepeer Agent-Tool-Profile: lean"
      ]
    }
  }
}

claude mcp add --transport http livepeer-agent \
  https://agent.livepeer.org/api/mcp \
  --header "Authorization: Bearer sk_YOUR_KEY" \
  --header "X-Livepeer Agent-Tool-Profile: lean"

{
  "name": "storyboard",
  "url": "https://agent.livepeer.org/api/mcp",
  "headers": {
    "Authorization": "Bearer sk_YOUR_KEY",
    "X-Livepeer Agent-Tool-Profile": "lean"
  }
}

codex mcp add --name storyboard \
  --url https://agent.livepeer.org/api/mcp \
  --header "Authorization: Bearer sk_YOUR_KEY" \
  --header "X-Livepeer Agent-Tool-Profile: lean"

{
  "mcpServers": {
    "livepeer-agent": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://agent.livepeer.org/api/mcp"]
    }
  }
}

Generate config locally: npm run mcp:config -- --client desktop --key sk_… or --demo

Install skills first

Health check (required)

1. "List available models and pricing."
2. "Generate a watercolor sunset over Tokyo — fast tier."
3. Confirm: agent echoes human_summary (not raw JSON) + image URL or inline image.

First prompt recipes

• Single image: "A watercolor sunset over Tokyo."
• 6-scene story: "Make me a 6-scene visual story about an octopus learning to ride a bike."
• Brand moodboard: "Build a 12-slot launch moodboard for [brand] — show cost first."
• Product placement: "Place this product in 6 scenes" (+ upload handoff)

Lean vs full tools

Hosts may show 9 recommended tools when STORYBOARD_LEAN_TOOLS=1 on the server: list_capabilities, get_pricing, create_media, generate_project, upload, get_cost_report, subscribe_progress, attach_skill, list_skills. Full surface (~99 tools) remains callable; send header X-Livepeer Agent-Tool-Profile: full for power users.

Naming the model you want

Pass model_override on create_media and the response tells you which model actually ran. When it matches, you get a confirmation. When it doesn't, you get the reason — because four different mechanisms can legitimately serve a different model than the one you named:

Exact dispatch — fail instead of falling back

When you'd rather see the failure than a substitute, use run_capability. It never fuzzy-matches: the name goes to the network verbatim, and it takes raw upstream params that create_media doesn't model (resolution, keyframe arrays, reference-image arrays, draft_cache_url).

run_capability({
  capability: "kling-o3-i2v",     // exact name, no substitution
  prompt: "a lantern drifting over dark water",
  source_url: "https://…/frame.jpg",
  params: { duration: 5, resolution: "1080p" }
})

Troubleshooting

• 401 — rotate key at app.daydream.live; check header spacing (see warning above).
• Tools show 0 — use mcp-remote shim (Desktop); quit fully and reopen.
• Desktop connector missing — see Desktop tab steps; verify identity is livepeer-agent via GET https://agent.livepeer.org/api/mcp.
• Large paste fails — use /upload-handoff or MCP upload / create_upload_url.
• Demo budget spent — add your Daydream key for unlimited renders.
• Stock clip marked done — read the leading STOCK SUBSTITUTE line; take Redo with AI.

← Back to /start · What's new · Model control · Templates · Playbooks