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.
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
The MCP endpoint is:
http://localhost:3000/mcp
Health/info endpoint:
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:
Validate the Worker bundle without publishing:
Deploy:
From the repo root, use:
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
| Environment variable | Default | Description |
|---|
PORT | 3000 | HTTP server port. |
WAVEDASH_DOCS_BASE_URL | https://docs.wavedash.com | Public docs origin. |
WAVEDASH_MCP_FETCH_TIMEOUT_MS | 10000 | Fetch timeout for docs requests. |
WAVEDASH_MCP_SEARCH_CACHE_MS | 300000 | Search index cache TTL. |
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.