Wavedash

Wavedash MCP

Read-only remote MCP server for Wavedash docs, SDK guidance, engine setup, and publishing checklists.

This server is intentionally unauthenticated. It only reads public docs from https://docs.wavedash.com and does not access user accounts, games, builds, local files, API keys, or private Wavedash data.

Tools

Four orthogonal tools, so clients rarely have to choose between overlapping options:

• wavedash_implementation_planner - plan and guide a Wavedash integration from the user's goal, engine, and requested features. The stage argument selects plan (architecture and reading order), setup (CLI install, sign-in, SDK init, wavedash.toml, local testing), features (SDK reference for the requested features), or deploy (upload and release checklist). Returns ordered steps, boundaries, and the relevant docs pages with content, as text and as structured output.
• search_wavedash_docs - search official Wavedash docs.
• get_wavedash_doc - fetch a docs page as Markdown.
• validate_wavedash_config - validate wavedash.toml text, including single-line or fenced input pasted from chat.

Engine, topic, and feature pages are resolved at runtime from the docs site's search index, so new docs pages need no change here.

Run Locally

npm installnpm start

The MCP endpoint is:

http://localhost:3000/mcp

Health/info endpoint:

http://localhost:3000/

Docker

docker build -t wavedash-mcp .docker run --rm -p 3000:3000 wavedash-mcp

Cloudflare Workers

The Worker entrypoint is src/worker.js, configured by wrangler.toml. It serves a stateless Streamable HTTP MCP endpoint at /mcp.

The Deploy MCP to Production GitHub Actions workflow deploys automatically on pushes to main that change MCP source, package dependencies, wrangler.toml, the root package.json, or the deployment workflow itself. Documentation and skill changes do not trigger deployment. Manual workflow runs remain available.

Local Worker dev:

npm run worker:dev

Validate the Worker bundle without publishing:

npm run deploy:dry-run

Deploy:

npm run deploy

From the repo root, use:

npm run mcp:deploy

The production MCP endpoint is:

https://mcp.wavedash.com/mcp

wrangler.toml declares mcp.wavedash.com as a Cloudflare Workers Custom Domain. If wavedash.com is in the same Cloudflare account, Wrangler can attach the hostname without a manual DNS record.

Remote Client Testing

Health/info endpoint:

https://mcp.wavedash.com/

Cloudflare AI Playground can test the deployed endpoint directly:

https://mcp.wavedash.com/mcp

Remote MCP clients that support Streamable HTTP can connect directly to the same endpoint.

For Claude Desktop or other local MCP clients that expect stdio, use mcp-remote:

{  "mcpServers": {    "wavedash": {      "command": "npx",      "args": ["mcp-remote", "https://mcp.wavedash.com/mcp"]    }  }}

Configuration

v0 / Lovable / Bolt Scope

Use this server as a no-auth Wavedash MCP listing. It can guide agents while they generate browser games, but it cannot upload or publish builds.

For authenticated steps, the MCP should instruct the human or host agent to use:

• Wavedash CLI install docs: https://docs.wavedash.com/cli/installation
• Wavedash CLI browser login: wavedash auth login
• CI/API token login: wavedash auth login --token ... or WAVEDASH_TOKEN
• Developer Portal manual upload/publish: https://wavedash.com/dev-portal

The Wavedash CLI is the command-line app for signing in, initializing wavedash.toml, testing locally, uploading builds, and publishing from a project folder. For users who do not want terminal commands, route them to the Developer Portal instead.

Do not pass API keys, session cookies, or private credentials as MCP tool arguments.

Authenticated tools such as creating games, uploading builds, or publishing should live in a separate authenticated MCP surface backed by Wavedash OAuth or scoped agent grants.