Skip to content

One-click deploys

The README has a deploy button for each hosted platform. Every button starts from the same public repository, so you can read exactly what gets deployed before you click.

Button Creates Good for
Deploy to Cloudflare The TypeScript hub as a Cloudflare Worker with a hibernating Durable Object A hub that costs $0 while idle
Deploy to Render The all-in-one container from render.yaml, with a disk at /data One service with the hub, a worker and search together
Deploy on Railway The all-in-one container from railway.json Same as Render, on Railway
Deploy on Deno The portable hub on Deno Deploy (hub/portable/deno-server.ts) A serverless hub off Cloudflare, with Turso and S3 for storage
Add to Home Assistant Adds the EdgeFetch add-on repository to Home Assistant OS A worker node on the box you already run

Fly.io has no button, but the repository includes fly.toml:

Terminal window
fly launch --copy-config
  1. Set two secrets. CALLER_TOKEN is what your agents and the /admin console use. AGENT_TOKEN is what worker nodes use to connect. Use long random values, for example openssl rand -hex 24.
  2. Open /admin on your new hub and sign in with the caller token.
  3. Connect your agent to https://<your-hub>/mcp. See Connect an MCP client.
  4. Add a worker if your hub has none built in. See Add worker nodes.

On Cloudflare, a hub with no workers still fetches through Cloudflare’s own network, but adding a worker at home gives you your own IP address and a headless browser for JavaScript pages.

All of these are off until you configure them:

Feature What to set
Web search SERPER_API_KEY, BRAVE_API_KEY or TAVILY_API_KEY, or SEARXNG_URL for a free self-hosted option
Stored artifacts and crawl history The ARTIFACTS R2 bucket and META D1 database in hub/wrangler.toml
Full-text search over what you’ve collected The SEARCH D1 database
Signed requests (Web Bot Auth) BOT_SIGNING_PRIVATE_KEY, from edgefetch keygen-bot-auth
Tracing OTEL_EXPORTER_OTLP_ENDPOINT and OTEL_EXPORTER_OTLP_HEADERS
Heavy single-page apps CRAWL4AI_API_URL, or FIRECRAWL_API_URL and FIRECRAWL_API_KEY, as a fallback after robots.txt passes