WeesaGen Hub

API Access

XVIEW Animator (embedded)

🤖 Connect your AI agent (MCP) — copy-paste setup

WeesaGen's tools are served to AI agents over MCP (Streamable HTTP) by the same Node server that hosts these pages — and your agent renders through this tab, so keep it open and active (background tabs freeze long renders). Setup is three steps plus one prompt you paste to your AI.

1 · Start the server (on the machine with the GPU browser)

cd server
npm install
npm start
# serves pages + API + MCP at http://localhost:8787  (change port: PORT=8790 npm start)
# needs Node 18+. Then open http://localhost:8787/weesagen-hub.html in Chrome/Edge (WebGPU).

2 · Claim an invite code on this page

1. Click Get API invite code above — the status bar turns Online.
2. Copy the 8-character code. Every generation-tool call passes it as code.
3. Codes expire after 24h — click Generate New Code for a fresh one.

3 · Register the MCP server in your agent (pick your client)

OpenCode — opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "weesagen": {
      "type": "remote",
      "url": "http://localhost:8787/mcp"
    }
  }
}

Claude Code — terminal:

claude mcp add --transport http weesagen http://localhost:8787/mcp

Cursor / Windsurf / Cline / Claude Desktop — MCP settings JSON:

{
  "mcpServers": {
    "weesagen": {
      "url": "http://localhost:8787/mcp"
    }
  }
}

Older Claude Desktop (stdio-only versions) — proxy bridge instead:

{
  "mcpServers": {
    "weesagen": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "http://localhost:8787/mcp"]
    }
  }
}

4 · Paste this prompt to your AI — it self-configures from here

Set up WeesaGen for me.

Context: WeesaGen is a set of browser-native AI tools (FLUX image generation,
vision chat, 3D Gaussian splats, text-to-speech, speech-to-text, music, a
Photoshop-style editor, XVIEW animation export, and a game Asset Library),
served over MCP (Streamable HTTP) by a local Node server and rendered through
a hub page I keep open in my browser.

Do this in order:
1. Register the MCP server in YOUR OWN client configuration if it is not there
   yet (Streamable-HTTP at http://localhost:8787/mcp, named "weesagen" — use
   your client's native way, e.g. an opencode.json "mcp" remote entry, or
   `claude mcp add --transport http weesagen http://localhost:8787/mcp`).
   If the server is not reachable, tell me to start it: `cd server`,
   `npm install`, `npm start` (Node 18+, default port 8787).
2. Ask me to open http://localhost:8787/weesagen-hub.html in Chrome/Edge,
   click "Get API invite code", keep the tab open and active, and paste the
   8-character code back to you.
3. Verify with the get_session_status tool using my code. If it reports
   offline, tell me to focus the hub tab and retry.
4. Then do whatever I ask using the weesagen tools, always passing my invite
   code as `code` on generation tools (generate_image, photoshop,
   xview_export, ...). The ingest_* library tools (ingest_search, ingest_post,
   ingest_download, ingest_analyze, ingest_review) need no renderer — they hit
   the online Asset Library directly. For long xview exports prefer
   frameStep 3+.

Good to know

1. Generation tools need this tab online — check first with get_session_status.
2. Library tools need no page at all. ingest_post uploads files from the machine running the server.
3. Serving 27 tools adds context — disable ones you don't need in your client if answers get bloated.