Overdare

An MCP server that lets AI agents (Claude Code / Claude Desktop) drive OVERDARE Studio — browse the DataModel, create and edit instances, write Luau scripts, and start playtests.

How it works

OVERDARE Studio already hosts a local JSON-RPC server (the same one its built-in "diligent" agent uses). This MCP server is a thin bridge to it — much like unity-mcp's connector, except Studio hosts the local server itself, so there's no plugin to install.

Claude  ──stdio──▶  overdare-mcp  ──HTTP POST──▶  Studio RPC (localhost:13377/rpc)
                                                      └─ OVERDARE Studio (must be running)

Studio must be open with a project for tools to work. If Studio isn't running, tools return a clear "cannot reach Studio" error instead of hanging.

Setup

npm install
npm run build

Verify the connection (open OVERDARE Studio first):

npm run probe                                   # smoke test + level.browse
npm run probe -- instance.read '{"path":"Workspace.Baseplate"}'   # inspect any method

Register with Claude Code

Add to .mcp.json in your project (an example is included in this repo):

{
  "mcpServers": {
    "overdare": {
      "command": "node",
      "args": ["/ABSOLUTE/PATH/TO/overdare-mcp/dist/index.js"]
    }
  }
}

For Claude Desktop, add the same block under mcpServers in claude_desktop_config.json.

Configuration (env vars)

Tools

Status / next steps

The RPC endpoint, port, and method names were derived from the OVERDARE Studio runtime. The exact param/response field shapes for each method still need to be confirmed against a live Studio — use npm run probe -- <method> '<json>' to capture real shapes, then tighten the Zod schemas in src/tools.ts. The overdare_rpc tool works regardless and is the safe fallback while iterating.