stick-director
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@stick-directorCreate a flipbook of a stick figure waving hello"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Stick Director
Stick Director is an MCP-powered stick-figure flipbook toy. A user describes an animation, connects an MCP agent, and watches the story, motion, frames, and quality review appear in the browser.
This repository contains the complete toy:
a Cloudflare Worker and MCP endpoint;
one MCP tool,
flipbook_step;temporary pairing sessions stored in Durable Objects;
schemas and validation for stories, motion, poses, and review;
SVG and JPEG renderers; and
a framework-free browser interface in
site/.
The Worker does not call a model. It returns structured requests through MCP, accepts the agent's answers, validates them, and advances the flipbook.
Read the companion article, You can just delegate things., for the idea behind the toy and its live prototype.
Try it locally
npm install
npm run dev # http://127.0.0.1:8787Open the page and click Wake it up. Add the generated MCP URL to an agent, then send the prompt shown on the page. The browser follows the session as the agent works and plays the accepted frames when the flipbook is ready.
The generated MCP URL controls one temporary session. Treat it like a password. The page also receives a read-only viewing URL and keeps the revoke URL in the browser that created the session. Sessions expire after 24 hours.
Related MCP server: Lousa
MCP flow
The agent uses the same tool for the entire run:
flipbook_step({ op: "start", goal, frameCount })
↓
judgement_required: story_plan | motion_plan | frame_batch | quality_review
↓
flipbook_step({ op: "answer", continuation..., answer })
↓
validate and advance, or return the next required answer
↓
complete: viewerUrlWhen more than one repair or next step is legal, the Worker can return the
technical routing request meta.next_judgement. Its answer selects one legal
contract and the useful context for it; it does not change flipbook state.
An answer call must echo the continuation values from the open request:
flipbook_step({
op: "answer",
continuationToken: "<continuation.token>",
requestId: "<continuation.requestId>",
requestFingerprint: "<continuation.requestFingerprint>",
answer: { /* payload matching request.program.outputSchema */ }
})Continue until the result is complete or blocked. Use inspect to read an
existing project when debugging:
flipbook_step({ op: "inspect", continuationToken: "<project id>" })Local viewer split
The Worker normally serves both the browser UI and API on port 8787. To test the same UI from a separate static origin:
npm run viewer # http://127.0.0.1:8788Keep npm run dev running. site/config.js points the static viewer back to
the Worker API on port 8787.
Deploy
Create the KV namespace used by the legacy direct MCP endpoint:
npx wrangler kv namespace create FLIPBOOKSPut its id in wrangler.jsonc. Confirm that the three rate-limit
namespace_id values are unused in the Cloudflare account, then deploy:
npm run deployWrangler uploads site/ as Worker Static Assets. The Durable Object binding
and its v1 SQLite migration are already declared.
The interactive page creates scoped MCP sessions at /mcp/s/<capability>.
The unscoped /mcp route remains for direct and legacy clients.
See INTERNALS.md for the implementation map.
This server cannot be deployed
Maintenance
Related MCP Connectors
Video review and collaboration for teams and agents: a full review interface for agentic video and motion graphics, feedback by voice, screen and drawing on the frame, agents that iterate version by version, and every version's notes and source files saved for the team.
Real-time collaborative whiteboard — AI agents and humans edit the same board live over MCP.
Make AI-drawn video from chat: your LLM draws every frame in code; add narration, music and SFX.
Durable agent-to-agent handoffs and shared scratchpad for multi-agent workflows.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceEnables AI agents to collaboratively draw and annotate Excalidraw diagrams in real-time via MCP tools, synced to a browser canvas.10Apache 2.0
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to draw animated, hand-drawn diagrams on a live whiteboard through MCP tools, allowing users to visually explain concepts and edit scenes interactively.16 npmAGPL 3.0
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to generate professional storyboards and videos from scripts or creative descriptions via MCP-compatible clients.9 npmMIT
- AlicenseNot gradedqualityCmaintenanceEnables LLM agents to generate labeled contact sheets of Manim animations, scaffold storyboards from renders, and validate rendered steps against expected intents via MCP tools.MIT