Skip to content

Protomaps is designed for simple CDN deployment, to turn your tilesets from these URLs:

To these URLs:{z}/{x}/{y}.mvt{z}/{x}/{y}.png{z}/{x}/{y}.mvt{z}/{x}/{y}.png

These installation guides should take less than one hour to deploy.


  • Serve tilesets to the web securely from private storage buckets.

  • Cache tiles at the edge for the fastest response to users, taking advantage of global CDN presence. Single-region buckets can have multi-second latencies for users far away; CDN cache hits can respond in 100 millisconds or less.

  • Backwards compatibility with existing map clients that use Z/X/Y map URLs.

  • Affordable costs at scale. There are no servers to maintain for AWS and Cloudflare deployments, and you pay nothing if there's no traffic.


A CDN deployment of Protomaps has three parts:

  • The data: a storage bucket containing PMTiles archives.

  • A serverless function that decodes incoming URLs into ranges and returns the tile bytes.

  • A edge network that caches the responses from serverless functions.

Next steps

Deployment Comparison Chart

Cloudflare is recommended for budget minded beginners. AWS is recommended for commercial companies who require a faster map and/or can offset increased marginal cost with monetization.

featurestatic pmtilesCloudflareAWSGCPCaddypmtiles serve
Z/X/Y compatible✖️
Edge caching✖️With CDNWith CDNWith CDN
SSLWith reverse proxy
Scale to zero$5 (USD)✖️✖️
Setup effort😓😓😓😓😓😓😓😓😓😓

Feature explanation

  • Z/X/Y compatible: Can tiles be requested as standard web mapping zoom, x, y tile coordinates?
  • Edge caching: Includes easy to configure edge network for content distribution (or manually paired with CDN) to achieve faster latency?
  • SSL: Supports encrypted link between tile server and client map library?
  • Scale to zero: Does the serverless function scale to zero cost during periods of low usage?
  • Setup effort: Developer time to configure complete cloud tile serving solution, less effort 😓 is better than more 😓😓😓 effort
  • Latency: Speedy maps tiles 🚀🚀🚀 load in ≤ 200 ms in the client for customers, slow tiles 🚀 load ≥ 500 ms
  • Cost: Total cost to run tile serving system, with 💰 being cheaper at $5 USD and 💰💰💰 more expensive options that including storage egress and/or CDN bandwidth costs to achieve lower latency

Coming soon

  • Azure support is in development and planned for 2024.

An open source mapping system released under the BSD and ODbL licenses.