wisewall-mcp

wisewall-mcp

One command puts your site live on the web. WiseWall actually hosts it — not a script bolted onto hosting elsewhere — and turns it into a real product on top: magic-link accounts, a per-member database, Polar-powered payments, and a server-side paywall. This package lets your coding agent drive all of it: deploy, configure a custom domain and DNS, manage secrets, create plans, without ever opening a dashboard. Works the same way with Claude Code, Claude Desktop, Cursor, Windsurf, VS Code (Copilot agent mode), Codex CLI, and Gemini CLI — or any other MCP-compatible client, since it's plain stdio MCP underneath.

No install step — npx runs it on demand. Nothing to sign up for beyond your email.

Deploy first, sign in later

One command, and it's live — skip the setup below entirely for a first look, no account at all:

npx -y wisewall-mcp deploy

No directory needed — it checks the current folder for a dist/build/out/public subfolder and deploys that, or the current folder itself if none exist (the common case for a plain static site with no build step). Pass one explicitly to override (deploy ./dist), and a name if you want one (deploy ./dist my-site or just deploy my-site to keep auto-detecting the folder). Prints a live preview URL immediately, plus a claim link. Click the claim link within 24 hours to add the site to your account (a magic-link email, same as Connect below) — otherwise it's deleted. Once you're ready to manage it from an agent (custom domain, plans, secrets), connect normally and use deploy/create_site as usual.

Already have a project instead of an empty folder? Run launch instead — same directory auto-detection/arguments as deploy, plus it scaffolds your coding agent's instructions file (CLAUDE.md and equivalents) with WiseWall's gating conventions first, printing a checklist of anything in the existing code worth double-checking. If your project has its own build step (a package.json with a build script), launch detects that and stops to tell you the exact command to run first, rather than deploying an empty or stale folder — it never runs a build itself. Re-running launch afterward picks up the result:

npx -y wisewall-mcp launch

Connect

npx -y wisewall-mcp login [e-mail]

Click the link that arrives by email — the command detects the click automatically and prints a token scoped to your own sites (never the platform's own credentials). It's shown once; copy it now.

Add it to your MCP client

The server config is the same everywhere — command: npx, args: ["-y", "wisewall-mcp"], and WISEWALL_TOKEN in the environment — only the file it goes in (and, for one client, the JSON shape) differs.

Claude Code — claude mcp add, or .mcp.json

claude mcp add wisewall --scope user --env WISEWALL_TOKEN=wisewall_... -- npx -y wisewall-mcp

See Bootstrapping from a brand-new Claude Code session below for a copy-pasteable one-liner that doesn't require having a token yet.

Claude Desktop — claude_desktop_config.json

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json (Settings → Developer → Edit Config also opens it)

{
  "mcpServers": {
    "wisewall": {
      "command": "npx",
      "args": ["-y", "wisewall-mcp"],
      "env": { "WISEWALL_TOKEN": "wisewall_..." }
    }
  }
}

Cursor — .cursor/mcp.json (project) or ~/.cursor/mcp.json (global)

{
  "mcpServers": {
    "wisewall": {
      "command": "npx",
      "args": ["-y", "wisewall-mcp"],
      "env": { "WISEWALL_TOKEN": "wisewall_..." }
    }
  }
}

Windsurf — ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "wisewall": {
      "command": "npx",
      "args": ["-y", "wisewall-mcp"],
      "env": { "WISEWALL_TOKEN": "wisewall_..." }
    }
  }
}

VS Code (Copilot agent mode) — .vscode/mcp.json

VS Code uses servers, not mcpServers, and each entry needs an explicit type:

{
  "servers": {
    "wisewall": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "wisewall-mcp"],
      "env": { "WISEWALL_TOKEN": "wisewall_..." }
    }
  }
}

Codex CLI — ~/.codex/config.toml (or project .codex/config.toml)

Codex's config is TOML, not JSON:

[mcp_servers.wisewall]
command = "npx"
args = ["-y", "wisewall-mcp"]
env = { WISEWALL_TOKEN = "wisewall_..." }

Gemini CLI — ~/.gemini/settings.json (or project .gemini/settings.json)

{
  "mcpServers": {
    "wisewall": {
      "command": "npx",
      "args": ["-y", "wisewall-mcp"],
      "env": { "WISEWALL_TOKEN": "wisewall_..." }
    }
  }
}

Get a token first with npx -y wisewall-mcp login [e-mail] (see Connect above) — then paste it in place of wisewall_....

Get your agent following WiseWall's conventions

Once connected, ask your agent to run the init tool (or run it yourself: it's exposed as one of wisewall-mcp's tools, callable from any of the clients above). It scaffolds a project-instructions file — CLAUDE.md, .cursor/rules/wisewall.mdc, .windsurf/rules/wisewall.md, .github/copilot-instructions.md, AGENTS.md, or GEMINI.md, depending on which client(s) you pass — with WiseWall's gating conventions (premium paths vs. sign-in-only paths, localStorage, secrets), so whichever agent works on this project next follows them from the first prompt instead of after the fact. Defaults to writing all six, idempotently — safe to run again later, and harmless for a client you don't actually use.

Already have an existing project instead of a fresh folder? Run convert instead — same scaffolding as init, plus a checklist of concrete things to go find and fix in that existing code (client-side-only paywall logic to remove, hardcoded API keys to move to set_secret, which paths actually need protectedPaths/authOnlyPaths) so it actually follows WiseWall's conventions, not just documents them.

Bootstrapping from a brand-new Claude Code session

Tell a session that knows nothing about WiseWall yet to "connect me to WiseWall," and give it just your email address — the one thing it can't guess on its own. It can do the rest itself, no cloning or building required:

npx -y wisewall-mcp login [e-mail]

Once the token prints (after you click the emailed link), it registers the server with Claude Code — ideally at user scope, so it's available in every project, not just this one:

claude mcp add wisewall --scope user \
  --env WISEWALL_TOKEN=<the token printed above> \
  -- npx -y wisewall-mcp

(Exact flags vary by Claude Code version — claude mcp add --help is the source of truth.)

From then on the connection persists: future sessions see the server already configured and discover the tools directly over MCP.

Pointing at a different WiseWall instance

Self-hosting, or targeting a test environment instead of platform.wisewall.net? Pass WISEWALL_API_URL explicitly, for both login and the server itself:

WISEWALL_API_URL=https://your-instance.example.com npx -y wisewall-mcp login [e-mail]

What it can do

27 tools, covering the same ground as the dashboard:

Category	Tools
Sites	list_sites, create_site (name + dist_dir), deploy (site_id + dist_dir), delete_site
Project setup	init (scaffolds this project's agent-instructions file(s) with WiseWall's conventions — see above), convert (same, for a project that already has code in it — also returns a compliance checklist)
Config	get_config, set_config
Domain / DNS	get_domain, set_domain, get_zone, create_dns_record, delete_dns_record, activate_custom_domain, redirect_www_to_root
Secrets	list_secrets, set_secret, delete_secret (values are never returned once written)
Plans	list_plans, create_plan, update_plan_paths, archive_plan
Members	list_members (read-only), export_members (bulk JSON export, read-only)
Tokens	list_tokens, create_token, revoke_token

create_site and deploy read a local folder (e.g. a build's ./dist) and upload it — nothing writes to platform storage directly. Every read-only tool is annotated readOnlyHint; every irreversible one (delete, archive) is destructiveHint, so a well-behaved MCP client can add extra confirmation on the actions that actually warrant it.

Security

A token only ever unlocks the sites on its own account. login proves you own the email address via a magic link sent by the platform itself — this command never asks for a password, and the platform never stores one to begin with.

Full product docs: wisewall.net · Dashboard: platform.wisewall.net