Rayberg MCP · connect your agent

Rayberg MCP

Drive Rayberg’s creative engine from any AI agent — on-brand images & video, powered by RAY’s knowledge and workflows. Your agent connects with a key, and everything it makes lands in your account.

Get your key How it connects

AlphaRAY — your director, live in the terminal

Get the Rayberg Studio

The creature, live in your terminal

macOS / Linux · install once

curl -fsSL https://rayberg.ai/cli/rayberg-studio.py -o rayberg-studio.py && python3 rayberg-studio.py install

Windows · install once

curl -fsSL https://rayberg.ai/cli/rayberg-studio.py -o rayberg-studio.py && python rayberg-studio.py install

One-time install (Python 3, no packages). It drops a rayberg command on your PATH — open a new terminal and just type rayberg. Add your rayberg_live_ key from Account → API keys (or the in-app login) for the live menus. On Windows use python (not python3).

Connect in 3 steps

From key to first render.

Rayberg speaks the Model Context Protocol over Streamable-HTTP. Point your agent at the endpoint, authenticate with your key, and Rayberg’s tools show up in the session.

Create your API key

Account → API keys

Mint a Rayberg key under Account → API keys. You’ll see it once — copy it right away. The key authenticates your agent as you: every image and clip it generates lands in your projects and bills to your balance.

Create your key

Add Rayberg to your agent

Streamable-HTTP MCP

Endpoint

https://rayberg.ai/api/mcp

Header Authorization: Bearer rayberg_live_YOURKEY — swap YOURKEY for the key from step 1.

Claude Code · CLI

claude mcp add --transport http rayberg https://rayberg.ai/api/mcp --header "Authorization: Bearer rayberg_live_YOURKEY"

Cursor.cursor/mcp.json

{
  "mcpServers": {
    "rayberg": {
      "url": "https://rayberg.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer rayberg_live_YOURKEY"
      }
    }
  }
}

Claude Desktopclaude_desktop_config.json

{
  "mcpServers": {
    "rayberg": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://rayberg.ai/api/mcp",
        "--header",
        "Authorization: Bearer rayberg_live_YOURKEY"
      ]
    }
  }
}

Generate

Tools appear in-session

Restart your agent. Rayberg’s tools appear in the session — ask it to explore models, read a knowledge pack, load a workflow, or generate. Images come back inline; video is async, returning a job id your agent polls with job_status.

Tools reference

What your agent can reach.

Ten tools, scoped to your account. Discovery and knowledge reads are free; generation bills tokens to your balance, one reservation per variation.

Discover

models_explore
Find the right model — recommend by goal, or get a model’s constraints. Filter by image/video and text/image input.
capability_get
The full structured spec for a model — inputs, modes, max refs, aspect ratios and exclusions. Read it before you generate.

Knowledge & workflows

get_knowledge
Read Rayberg’s craft knowledge shelf — the same packs RAY uses (models, cinema, story, marketing, formats).
list_workflows
List the multi-step production playbooks — explainer video, viral hook ad, commercial, shorts restyle.
load_workflow
Load one playbook’s ordered phases, cost shape and done-criteria for your agent to follow.
chat_with_ray
Talk to RAY directly — ask for direction, a look or the next step, and get answers grounded in the studio’s knowledge.

Generate

generate_image
Render an image into your account — cinematic, photoreal, reference-locked — billed to your balance.
generate_video
Render a video (async) into your account — Seedance, Kling, Veo — with native audio where supported.

Projects & status

list_projects
List your Rayberg projects (workspaces) so you can target a render, or let a default workspace be used.
job_status
Check an async job — poll for a terminal state, or wait synchronously up to ~25s for the result.

Command-line native

One command. Full studio.

The same engine, spoken from your terminal. The Rayberg CLI drives the studio over the very same MCP — one command generates on-brand, waits for the render, and lands the result in your account. Here's the vibe.

rayberg — zsh — 96×24MCP · live

RAYBERG

✦ The creative supercomputer — in your terminal.

✦rendering—0.0s

Coherent by design

One look, not a random AI mix.

Ask for a whole campaign or a full website of assets and the pieces should feel like one production — not nine different generators fighting each other. Rayberg holds the look with RAY’s knowledge shelf and its production playbooks: a scene’s references and style stay locked across every shot, so characters, palette and grade carry through.

Call list_workflows to see the playbooks, then load_workflow for the ordered phases — your agent follows them with generate_image and generate_video, so the whole set holds together.

Reference lockStyle continuityAssembled cut, not loose clips

Production playbooks

Explainer videoexplainer-video
Research → narrated ~10s blocks → one clip per block → assembled cut.
Viral hook adviral-hook-ad
A scroll-stopping UGC hook: interrupt → unbothered → pivot into the point.
Shorts restyleshorts-restyle
Existing video → 10s beats → every beat re-rendered in one preset look → re-cut short.
Commercial productioncommercial-production
One concept → the whole spot planned → reference truth built → every cut bound to it → assembled commercial.

Honest about the alpha

Alpha

The Rayberg MCP is an early alpha. It’s fail-closed — the endpoint stays off unless it’s enabled on the deployment, and unauthenticated calls are rejected. Tools, names and shapes can still change as we harden it. Keys are yours to revoke any time under Account → API keys, and generation runs against your own balance with the same billing rules as the studio. Tell us what breaks: hello@rayberg.ai.

Get your key See token pricing

From key to first render.

Rayberg speaks the Model Context Protocol over Streamable-HTTP. Point your agent at the endpoint, authenticate with your key, and Rayberg’s tools show up in the session.

Create your API key

Account → API keys

Create your key

Add Rayberg to your agent

Streamable-HTTP MCP

Endpoint

https://rayberg.ai/api/mcp

Header Authorization: Bearer rayberg_live_YOURKEY — swap YOURKEY for the key from step 1.

Claude Code · CLI

claude mcp add --transport http rayberg https://rayberg.ai/api/mcp --header "Authorization: Bearer rayberg_live_YOURKEY"

Cursor.cursor/mcp.json

{
  "mcpServers": {
    "rayberg": {
      "url": "https://rayberg.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer rayberg_live_YOURKEY"
      }
    }
  }
}

Claude Desktopclaude_desktop_config.json

{
  "mcpServers": {
    "rayberg": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://rayberg.ai/api/mcp",
        "--header",
        "Authorization: Bearer rayberg_live_YOURKEY"
      ]
    }
  }
}

Generate

Tools appear in-session

What your agent can reach.

Ten tools, scoped to your account. Discovery and knowledge reads are free; generation bills tokens to your balance, one reservation per variation.

Discover

models_explore
Find the right model — recommend by goal, or get a model’s constraints. Filter by image/video and text/image input.
capability_get
The full structured spec for a model — inputs, modes, max refs, aspect ratios and exclusions. Read it before you generate.

Knowledge & workflows

get_knowledge
Read Rayberg’s craft knowledge shelf — the same packs RAY uses (models, cinema, story, marketing, formats).
list_workflows
List the multi-step production playbooks — explainer video, viral hook ad, commercial, shorts restyle.
load_workflow
Load one playbook’s ordered phases, cost shape and done-criteria for your agent to follow.
chat_with_ray
Talk to RAY directly — ask for direction, a look or the next step, and get answers grounded in the studio’s knowledge.

Generate

generate_image
Render an image into your account — cinematic, photoreal, reference-locked — billed to your balance.
generate_video
Render a video (async) into your account — Seedance, Kling, Veo — with native audio where supported.

Projects & status

list_projects
List your Rayberg projects (workspaces) so you can target a render, or let a default workspace be used.
job_status
Check an async job — poll for a terminal state, or wait synchronously up to ~25s for the result.

One command. Full studio.

rayberg — zsh — 96×24MCP · live

RAYBERG

✦ The creative supercomputer — in your terminal.

✦rendering—0.0s

One look, not a random AI mix.

Call list_workflows to see the playbooks, then load_workflow for the ordered phases — your agent follows them with generate_image and generate_video, so the whole set holds together.

Reference lockStyle continuityAssembled cut, not loose clips