Discover Scribe tools
A compatible client detects the Site tools available in the open editor. The current validated Desktop build exposes 41 tools, but the exact surface can evolve.
Experimental • AI-assisted editing • WebMCP Site tools
Let AI agents work inside the real Scribe project.
Compatible browser agents can discover semantic Scribe Site tools, inspect the project you can see, and make intentional edits without relying on fragile screen coordinates or creating a hidden second editor.
Why it matters
AI assistance works better when it understands editor actions.
Screen-click automation guesses where controls are. WebMCP publishes typed, task-level actions instead. A compatible agent can search Scribe's capabilities, inspect the exact input contract, and execute the canonical operation against the visible project.
Current-build discovery evidence
See Scribe’s Site tools before an agent acts.
The compatible client associates the tools with app.scribeanimator.com and shows the current read/write surface. This complements the visible edit result above: discovery first, then an intentional operation in the real editor.
How it works
Semantic help, with Scribe's normal controls still authoritative
Start read-only
Ask the agent to inspect project state, scenes, or locked objects first. Early discovery and status remain available while project continuity initializes.
Make a reversible edit
Agent actions update the same project shown on the canvas. Where the underlying Scribe capability supports history, ordinary Undo remains available.
Discover specialized actions
For larger tasks, the agent searches capabilities, obtains the exact schema, then executes that canonical capability through the broker.
Verify on the canvas
Review the visible scene, timeline, selection, or inspector just as you would after a manual change. Locks, permissions, entitlements, and validation still apply.
Keep protected choices human
The agent can prepare workflows, but local-file selection, recording-source permission, credentials, purchases, and similar browser-owned decisions remain yours.
Try asking
Natural-language tasks, grounded in real capabilities
“Inspect the current project and list locked objects.”
“Create a new scene, add a heading, center it and animate its entrance.”
“Find a drawing that fits this explanation and add it to the scene.”
“Move this narrative beat three seconds later.”
Designed-in boundaries
The compatible client supplies the intelligence.
- Scribe does not host an LLM for WebMCP; the compatible browser client or agent supplies the AI.
- WebMCP and OpenAI Site tools availability varies by client version, rollout, workspace, browser, and model.
- The agent does not bypass authentication, project permissions, plan limits, object locks, or editor validation.
- Protected browser and account surfaces remain human-controlled, including file choosers, screen-recording sources, credentials, and purchases.
- Public share and embed pages do not receive the private editor tool surface.
Keep going