Skip to main content
Version: Current

Use Scribe Animator with ChatGPT/Codex Site tools

WebMCP gives a compatible browser agent typed, semantic ways to work with the Scribe project already open on your screen. In clients that expose this integration as Site tools, you can ask in ordinary language; you do not need to memorize internal tool names.

One visible project

Agent actions operate on the real project in the Scribe editor. There is no hidden second editor. Normal authentication, permissions, plan limits, locks, validation, and history remain authoritative.

What you need​

Recreating a reference image or diagram? Use the dedicated reference reconstruction workflow before importing artwork or animating the scene.

  • Scribe Animator open in its Desktop web editor;
  • a compatible browser client that implements document.modelContext;
  • Site tools enabled and available for the current client/session; and
  • a disposable or safely checkpointed project for your first edit.

1. Open Scribe in a WebMCP-compatible client​

Open Scribe Animator in the compatible client's browser. Wait for the project to load completely; the early discovery/status surface is available first, then the full catalog becomes available after project continuity initializes.

2. Open Site tools​

Open the client's Site tools or website-tools control. In ChatGPT or Codex, the exact location can vary with the current client. Confirm that the tools are associated with app.scribeanimator.com, not with an unrelated tab.

3. Confirm Scribe tools are visible​

Check the tools listed in your current session. The capture below recorded 41 Site tools in an earlier verified session; it does not establish the count or availability in your client today. What matters is that the Scribe status/discovery tool and the continuity- protected editing catalog are available.

Site tools panel for app.scribeanimator.com showing 41 available Scribe tools, including 17 read and 24 write tools, over the visible editor
Recorded discovery evidence: this session listed 41 tools—17 read and 24 write—for app.scribeanimator.com. Inspect your current session for its actual tools and availability.
Screenshot callouts
  1. Scribe editor site
    The client associates this tool surface with app.scribeanimator.com.
  2. Recorded tool count
    This captured session exposed 41 tools: 17 read and 24 write.
  3. Discoverable Scribe tools
    The panel exposes status, capability discovery, schema inspection, execution, context, diagnostics, and automation entry points.

4. Ask a read-only question​

Begin with:

Inspect the current project and list locked objects.

The answer should describe the open project's real state without changing it. If the client cannot see the project, check the troubleshooting section before asking for edits.

Before a larger task, ask the agent to inspect project.get_status, create a project.save_local checkpoint, and retain the returned project ID and checkpoint key. A local checkpoint is a recovery point in this browser profile; use a .scribe backup for a portable copy.

5. Ask for one reversible edit​

Try:

Create a new scene, add “Hello WebMCP”, center it and animate its entrance.

The agent can combine scene creation, text insertion, layout, and animation capabilities. Where the underlying capability supports Scribe history, the change remains available to normal Undo.

6. Verify the visible result​

Inspect the canvas, scene list, inspector, and timeline. The visible Scribe project is the result—there is no terminal-only clone to reconcile later.

A real Scribe Animator scene created through WebMCP Site tools, with search, inspect, and execute concepts visible on the canvas
Real current-build evidence: semantic Site tools created and animated content in the visible Scribe project.
Screenshot callouts
  1. Visible agent-edited scene
    The result is ordinary editable content in the real Scribe canvas.
  2. Semantic capability workflow
    The scene summarizes the search, schema inspection, and canonical execution model.
  3. Normal editor timeline
    Agent-authored objects participate in the same scene timing and history as manual edits.

7. Undo the edit​

Use Scribe's normal Undo control. Confirm that the reversible changes are removed from the same visible project. Some operations are inherently read-only or have their own confirmation boundary; the agent cannot turn every action into an undoable edit.

8. Ask for a multi-step task​

Once the small edit is verified, try one of these supported workflows:

Find an appropriate drawing for this scene and insert it.

Move this StoryBeat three seconds later.

Be precise about the target scene, selection, or StoryBeat when several could match. The agent should inspect current state before making a consequential change.

For an original explainer, try:

Create an editable whiteboard explainer about this topic. Plan the story beats, establish readable static scenes, add native relationships and purposeful motion, preview event frames and camera holds, then play the complete result. Report what was checked and the actual save result.

The Scribe Animator Director workflow expands this production method and explains the separate prerequisites for the versioned skills package.

9. Use capability discovery for specialized tasks​

Scribe exposes a broad canonical capability catalog through a stable broker workflow. A compatible agent can:

  1. search for relevant capabilities;
  2. inspect the exact schema and current availability; and
  3. execute the selected canonical capability.

Natural-language users do not need to type those steps manually. Asking the agent to “find the supported Scribe workflow first” is enough for most tasks.

Advanced: the broker tools​

  • scribe_get_webmcp_status reports sanitized readiness and continuity state.
  • scribe_search_capabilities finds task-relevant actions.
  • scribe_get_capability returns the exact input schema and availability.
  • scribe_execute_capability runs the canonical capability after validation.

The broad catalog remains reachable without registering every capability as a separate browser tool.

Check a chosen frame​

Ask the agent to discover ui.project_frame_preview. In the current contract, action: "show", the returned sceneId, and scene-local timeMs display a frame at the project's pixel dimensions. displayMode: "fit" fits the frame into the overlay; "actual" shows it at its actual CSS pixel size. Use Close, Escape, or action: "hide" to dismiss it.

The result reports frame metadata after the preview is displayed. Inspect the visible frame to judge typography, drawing, connectors, and camera framing; the tool does not return image bytes. A failed or cancelled preview is an unverified frame. The preview preserves the authored project, history, selection, and playhead. Follow it with live playback to check motion.

10. Understand human-controlled browser actions​

WebMCP does not bypass protected choices. You still choose or approve local files, the tab/window/screen used for recording, microphone permission, credentials and OTPs, purchases, destructive account confirmations, and browser-owned download or share destinations.

Scribe does not host an LLM for WebMCP. The compatible client supplies the intelligence; Scribe supplies the semantic editor tools and enforces its normal product rules.

SVG export is unavailable in the current public editor and WebMCP API. SVG import and drawing remain available. For delivery, discover video export or use a .scribe package for editable backup.

11. Troubleshooting​

No Scribe Site tools appear: confirm that the client supports top-level imperative WebMCP tools, the Scribe editor is the active page, and document.modelContext exists. A normal extension alone is not proof of support.

Status is ready but an edit is unavailable: wait for the project to finish loading, then ask the agent to inspect the capability's availability. Some actions need an active scene, selection, media object, or signed-in entitlement.

A task stops at a chooser: complete the browser-owned file, recording, permission, or purchase decision yourself, then ask the agent to continue.

The result targeted the wrong content: Undo, name the intended scene or object more specifically, and retry with a read-only inspection first.

The preview fails or the project changes while rendering: recheck project identity and capability availability, resolve missing media, and request the frame again. Do not treat a blank or stale overlay as a successful check.

Verification note​

Reviewed on 7 October 2026 against the current Scribe capability contracts and regression coverage. The figures are recorded real-client captures. They do not establish a fresh Director installation, today's client rollout, or production deployment of every capability described here.

Last updated on