AI Agents, CLI & MCP

Build worlds with
your AI agent.

Ask the AI assistant built into your dashboard, or bring your own agent. Allegory has a command line and an MCP server, so Claude, ChatGPT or OpenClaw can list your scenes, edit entities and push scene code.

Works with Claude ChatGPT OpenClaw + any assistant that writes HTML

Available now

Four ways AI works on your scenes.

The built-in assistant, the terminal and your own agent all work on the same scenes — and the activity log shows which door each change came through.

Built-in AI assistant

Available

Describe what you want in your Allegory dashboard and the assistant builds it. Nothing to install or connect.

Try it free →

Allegory CLI

Available

Scenes, entities, components and scripts from any terminal. Table output for people, --json for scripts and agents.

Install →

MCP server

Available

One tool per CLI command at app.allegoryxr.com/mcp. Connect Claude Code, Claude Desktop or claude.ai in a minute.

Connect →

Your own editor

Available

Pull a scene’s HTML, JavaScript and CSS into a folder, edit with your tools, push back what changed.

How it works →

Command line

allegory, in your terminal.

Commands are allegory <resource> <action>. The list comes from Allegory itself, so new commands appear without updating the CLI — allegory help always shows what your account can run today.

  • Sign-in works over ssh and inside containers
  • --json prints the raw result; errors come back as JSON too
  • Pick an account once with allegory use "Acme Studio"
Terminal
# install — Node 20 or later, no dependencies
npm install -g allegoryxr

# sign in: prints a link and a code — approve in any browser
allegory login

allegory scenes list --search mars
allegory scenes show 42
allegory entities update 7 --name "Airlock"
allegory components add --entity 7 --type light
allegory activity list --limit 10

# everything, grouped by resource, with examples
allegory help
No dependencies · no telemetry

Scene code

Edit scene scripts in your own editor.

Pull a scene’s scripts into a folder, one file each, and push back only what changed. Great for you — and for coding agents that live in a repo.

  • If someone changed a script on the server since you pulled, push skips it and tells you
  • --dry-run shows the plan; --json reports every file for CI
  • New .js, .css or .html files become new scripts
mars-base/
# writes ./mars-base/, one file per script
allegory scripts pull scene_abc123
cd mars-base && $EDITOR spin.js

# what would be sent, then send what changed
allegory scripts push --dry-run
allegory scripts push

# new file? attach it to an entity as it goes up
allegory scripts push --attach wobble.js=12
Deleting a file never deletes the script — push lists it instead

MCP server

Connect your agent in one line.

Allegory is an MCP server with one tool per CLI command. Your agent signs in through the browser with OAuth — no API keys to paste.

  • Claude Desktop and claude.ai: Settings → Connectors → Add custom connector, with https://app.allegoryxr.com/mcp
  • Stdio-only clients: npx -y mcp-remote https://app.allegoryxr.com/mcp
  • Run allegory mcp to print these instructions any time
Claude Code
# Claude Code
claude mcp add --transport http allegory \
  https://app.allegoryxr.com/mcp

# or a project's .mcp.json
{
  "mcpServers": {
    "allegory": {
      "type": "http",
      "url": "https://app.allegoryxr.com/mcp"
    }
  }
}
Streamable HTTP · OAuth sign-in in the browser

Built for agents

Safe to hand to an agent.

Deletes need a yes

A delete shows what it removes and asks you to type its name. Without a terminal — in a script or agent — it is refused unless --yes is given.

Roles respected

Any member can read and change things; deleting needs the account’s owner or an admin. The CLI says so before sending.

Activity log

Who changed what, and through which door: the editor, AI, the CLI or MCP.

Predictable output

--json on stdout, JSON errors on stderr, and exit codes 0, 1 and 2 an agent can act on.

Example

One prompt, one component, one command.

Ask any assistant for an A-Frame component, save it, and put it on an entity from the terminal.

  • Works with Claude, ChatGPT, OpenClaw and other assistants
  • The same component runs on a headset, a phone and a laptop
  • Any text argument can come from a file: --code-file, --attributes-file
spin.js

Write an A-Frame component that slowly spins an entity. Make the speed configurable.

Claude
AFRAME.registerComponent('spin', {
  schema: { speed: { default: 20 } },
  tick(time, delta) {
    const deg = this.data.speed * delta / 1000;
    this.el.object3D.rotation.y +=
      THREE.MathUtils.degToRad(deg);
  }
});
Then: allegory entities update 12 --code-file spin.js

Questions

Agents, answered.

How do I install the CLI?

You need Node 20 or later. Run npm install -g allegoryxr, then allegory login and approve the sign-in in your browser. allegory help lists everything you can run.

Is there AI built into Allegory?

Yes. Your Allegory dashboard has a built-in AI assistant: describe what you want and it works on your scene. Changes it makes appear in the activity log, marked AI. You can also bring your own agent through the CLI or MCP.

What can an agent do?

Whatever the CLI can: list and create scenes, show a scene’s entities, update entities and components, pull and push scene scripts, and read the activity log. Run allegory help for the current list.

Which assistants work?

Any MCP client can connect to https://app.allegoryxr.com/mcp — Claude Code, Claude Desktop and claude.ai among them. Coding agents that run terminal commands can use the CLI directly.

Can an agent delete my scenes?

Only if its account role allows deletes (owner or admin), and only with an explicit --yes when there is no person at a terminal to confirm. Every change shows up in the activity log.

Is Allegory affiliated with Anthropic or OpenAI?

No. Claude, ChatGPT and OpenClaw are named because they work with A-Frame and MCP. Their names and logos belong to their owners.

Build your first world today.

Start free — no card required.