Self-hosting wmcp.sh

Fork the repo and run your own instance on Cloudflare Workers — either as your team's private MCP gateway, or as a fully-branded fork. Both worker code (TypeScript) and adapters (ES modules) are MIT-licensed.

This guide assumes you've already got a Cloudflare account and wrangler CLI logged in.

1. Fork + clone

gh repo fork New1Direction/webmcp-anything --clone
cd webmcp-anything/worker
npm install

2. Create your KV namespaces

./node_modules/.bin/wrangler kv:namespace create CACHE
./node_modules/.bin/wrangler kv:namespace create KEYS
./node_modules/.bin/wrangler kv:namespace create USAGE

Each command prints the new namespace ID. Update wrangler.toml:

[[kv_namespaces]]
binding = "CACHE"
id = "<your-CACHE-id>"

[[kv_namespaces]]
binding = "KEYS"
id = "<your-KEYS-id>"

[[kv_namespaces]]
binding = "USAGE"
id = "<your-USAGE-id>"

3. Set secrets

# Required
echo "<random 64-hex>" | ./node_modules/.bin/wrangler secret put ADMIN_TOKEN
node -e "console.log(require('crypto').randomBytes(32).toString('hex'))" | ./node_modules/.bin/wrangler secret put TOKEN_ENC_KEY

# OAuth providers you want to support
echo "<gh-client-id>"     | ./node_modules/.bin/wrangler secret put GITHUB_CLIENT_ID
echo "<gh-client-secret>" | ./node_modules/.bin/wrangler secret put GITHUB_CLIENT_SECRET
# ...repeat for GOOGLE_*, SLACK_*, NOTION_*, LINEAR_*, DISCORD_*, STRIPE_*, ANTHROPIC_API_KEY (for the LLM-fallback adapter), etc.

The full secret list lives in worker/src/env.d.ts.

⚠️ If you lose TOKEN_ENC_KEY, every stored OAuth token in KV becomes unreadable. Treat it like a database backup key — store in 1Password or your KMS.

4. Optional: set your own domain

./node_modules/.bin/wrangler deploy
# By default this deploys to <worker-name>.<account>.workers.dev

# Then add a custom domain (Cloudflare dashboard):
#   Workers & Pages → your worker → Settings → Triggers → Custom Domains → Add Custom Domain
#   Enter: yourbrand.com

Cloudflare provisions the cert + DNS automatically. Custom Domain is preferred over Worker Route — Routes don't auto-provision certs and have gotchas with caching.

5. Customize branding

The brand identity is intentionally light-touch — change these to make it yours:

File	What
worker/src/landing.ts	Homepage hero, copy, CTAs
worker/src/og.ts	OG image SVG generator
`worker/public/og.png$	\text{Static} \text{fallback} \text{OG} \text{image} (1200 \times 630)
$worker/src/blog.ts BRAND_CSS`	Blog brand colors (default: dark purple/cyan)
worker/src/u.ts llmsTxt()	/llms.txt navigation copy

CSS variable names are consistent across pages — change one set and it propagates:

:root {
  --bg:      #07070d;
  --card:    #16161f;
  --bg2:     #11111c;
  --border:  #26263a;
  --text:    #ececf5;
  --muted:   #8a8aa8;
  --accent:  #7c5cff;  /* primary brand */
  --accent2: #00e5ff;  /* secondary brand */
}

6. Optional: customize adapters

The adapters/ directory is the shared adapter library — same files run in both the worker and the Chrome extension. To add an adapter for a site we don't yet support:

cp -r adapters/_template adapters/yoursite
# edit adapters/yoursite/adapter.js — implement detect() + extract()
node adapters/_test/run.mjs

See adapters/CONTRACT.md for the interface.

7. Deploy

./node_modules/.bin/wrangler deploy

A typical full-deploy is ~5 seconds — CF builds, uploads (~1.2 MB compressed), and rotates traffic.

To deploy in stages, use --env staging (you'll need a [env.staging] section in wrangler.toml).

8. Cron

The default wrangler.toml ships with cron: 0 */2 * * * — re-seeds the directory + refreshes stale tool caches every 2 hours. The handler is at worker/src/cron.ts. You can change the schedule or remove the cron block entirely.

9. Optional: GitHub OAuth for /dashboard sign-in

If you want users to sign in to your fork via GitHub:

1. https://github.com/settings/developers → New OAuth App
2. Homepage URL: https://yourbrand.com
3. Authorization callback URL: https://yourbrand.com/api/v1/auth/github/callback
4. Copy Client ID + secret into wrangler secrets (GITHUB_CLIENT_ID, GITHUB_CLIENT_SECRET)

Repeat the same pattern for each provider in worker/src/providers.ts.

10. Optional: Stripe billing

If you want self-serve Pro/Enterprise billing:

1. Stripe dashboard → create products + prices ($999/mo Pro, $4,999/mo Enterprise — or your own pricing)
2. Add Price IDs to worker/src/stripe.ts
3. Create a webhook endpoint pointing at https://yourbrand.com/api/v1/stripe/webhook
4. Set secrets: STRIPE_SECRET_KEY, STRIPE_WEBHOOK_SECRET, STRIPE_PRICE_PRO, STRIPE_PRICE_ENTERPRISE

What you get for free

• Sub-50ms edge response anywhere in the world (CF's ~300 POPs)
• KV-backed caching with no extra cost for the first 100K reads/day
• Auto SSL + DDoS via CF
• 1.2 MB worker bundle ships in <2s deploys
• 5-tier adapter chain works out of the box for ~4M+ Shopify storefronts + every site with JSON-LD or OpenAPI

What you have to bring

• An Anthropic API key (ANTHROPIC_API_KEY secret) if you want the LLM-fallback adapter active
• OAuth app credentials per provider (the worker won't proxy a provider without its client_id + secret)
• A domain (optional but recommended — *.workers.dev URLs look weird in MCP configs)

What you should NOT change without thinking

• TOKEN_ENC_KEY after launch — every stored OAuth token in KV becomes unreadable
• The KV key prefix patterns in worker/src/directory_capture.ts / directory_admin.ts — dirsub:, lead:, verified:, featured: are load-bearing across the admin endpoints
• The reverse-ts sort pattern — switching to forward-ts breaks the newest-first guarantee on KV.list({prefix})

---

Questions? Open an issue at github.com/New1Direction/webmcp-anything/issues or check the original deployment at wmcp.sh.