Maplibre Pmtiles Patterns logo

Maplibre Pmtiles Patterns

Organization
maplibre
maplibre-pmtiles-patterns

Serverless vector and raster tiles with PMTiles for MapLibre GL JS — single-file format, HTTP range requests, hosting on S3/R2/GitHub Pages, generating with Planetiler or tippecanoe, and the pmtiles protocol. Use when you need no tile server or want to host tiles from static storage.

Overview

Publishermaplibre
Repositorymaplibre-agent-skills
Skill namemaplibre-pmtiles-patterns
Stars
148
Forks
10
Bundled files
Instructions only
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by maplibre on GitHub. Read the source before you install it.

Installation

Install the Maplibre Pmtiles Patterns AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/maplibre/maplibre-agent-skills.git /tmp/maplibre-agent-skills
mkdir -p .claude/skills
cp -r /tmp/maplibre-agent-skills/skills/maplibre-pmtiles-patterns .claude/skills/maplibre-pmtiles-patterns
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Maplibre Pmtiles Patterns in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Maplibre Pmtiles Patterns on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Maplibre Pmtiles Patterns is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

MapLibre PMTiles Patterns

PMTiles is a single-file format for vector or raster map tiles. You host one (or a few) files on any static host; MapLibre requests byte ranges over HTTP. No tile server, no dynamic backend. This skill covers when to use PMTiles, how to generate and host them, and how to connect them to MapLibre GL JS.

When to Use This Skill

  • Hosting map tiles without running a tile server (S3, Cloudflare R2, GitHub Pages, etc.)
  • Building a fully static or serverless map stack
  • Serving large tile sets from a CDN with range requests
  • Generating PMTiles from OSM or other sources (Planetiler, tippecanoe)
  • Using Overture Maps or other single-file tile datasets with MapLibre

What PMTiles Is and Why It Matters

  • Vector and raster — PMTiles supports both. A file can contain vector layers (e.g. water, roads, POIs), raster imagery (PNG/JPEG), or raster-dem (elevation, e.g. Terrarium format for terrain). In the style you use type: 'vector', type: 'raster', or type: 'raster-dem' accordingly.
  • Single file per map — One .pmtiles file typically contains the full tile pyramid (all zoom levels) and all layers (vector or raster) in one archive. The format stores tiles in a compact layout (e.g. Hilbert curve) so the client can request only the byte ranges it needs. For very large coverage you may split by region into multiple files.
  • HTTP range requests — The client requests only the byte ranges it needs (e.g. one tile), so the server does not need to understand x/y/z. Any host that supports Range headers works.
  • Serving — You can serve directly from static storage (S3, R2, GitHub Pages, Netlify): the client uses range requests, so no tile server is required. Alternatively, tileserver-gl or Martin can serve PMTiles (from local paths, HTTP URLs, or S3), useful if you want one server that also provides styles, glyphs, or other sources.
  • Creating — You can get PMTiles by converting from MBTiles (PMTiles CLI) or by generating from source data (Planetiler, tippecanoe, GDAL, etc.). Alternatively, Protomaps is a provider where you can download pre-built PMTiles (e.g. global or regional basemaps) and serve them yourself, or create custom extracts via the PMTiles CLI—no need to generate from OSM yourself. Protomaps basemaps are built from OpenStreetMap data; OSM attribution is required in any map that uses them. See The PMTiles CLI and Generating PMTiles below.
  • Good for CDNs — Range requests cache well; put the file behind a CDN for fast global access.

When to prefer PMTiles over a traditional tile server:

  • You want zero server logic (static hosting only).
  • You have a bounded dataset (country, region, theme) that fits in one or a few files.
  • You want simple deployment and low ops (upload file, set cache headers, done).

When to prefer a tile server (e.g. tileserver-gl, Martin):

  • You need dynamic tiles from a database (PostGIS) or frequently updated data.
  • You have a very large global dataset and want to generate tiles on demand or by region only.

MapLibre Integration: The PMTiles Protocol

MapLibre does not speak PMTiles natively. You use the PMTiles library to add a protocol handler so that a pmtiles:// (or https:// to a .pmtiles file) source works.

Install:

bash
npm install pmtiles

Register the protocol and use in a style:

javascript
import * as pmtiles from 'pmtiles';
import * as maplibregl from 'maplibre-gl';
import 'maplibre-gl/dist/maplibre-gl.css';

// Add PMTiles protocol so sources can reference .pmtiles URLs
const protocol = new pmtiles.Protocol();
maplibregl.addProtocol('pmtiles', protocol.tile);

const map = new maplibregl.Map({
  container: 'map',
  style: {
    version: 8,
    sources: {
      tiles: {
        type: 'vector',
        url: 'pmtiles://https://example.com/data.pmtiles'
      }
    },
    layers: [
      {
        id: 'background',
        type: 'background',
        paint: { 'background-color': '#f8f4f0' }
      },
      {
        id: 'water',
        type: 'fill',
        source: 'tiles',
        'source-layer': 'water',
        paint: { 'fill-color': '#a0c8f0' }
      }
      // add more layers as needed — each uses the same source, different 'source-layer'
    ]
  },
  center: [0, 0],
  zoom: 2
});

// Optional: remove protocol on map teardown
// map.on('remove', () => maplibregl.removeProtocol('pmtiles'));

addProtocol changed in MapLibre GL JS 4.0.0. The v3 (params, callback) signature is gone — the 4.0.0 changelog entry reads "Changes addProtocol to be promise-based without the usage of callbacks and cancelable". A handler is now async (params, abortController) and must resolve to an object with a data property: {data, cacheControl?, expires?}. Resolving to a bare ArrayBuffer, Blob, Response or Uint8Array is not the contract — MapLibre reads .data off whatever the handler resolves with, so a bare buffer reads as no data at all and an arrayBuffer request is filled with an empty ArrayBuffer(0): the tile comes back blank instead of erroring. Register the handler on the module — maplibregl.addProtocol(...), or the named addProtocol import — once, before the first map is constructed: the protocol registry is global, so one registration covers every map on the page.

js
// ❌ v3 — on v4+ the second argument is an AbortController, so callback(...) throws,
// and the returned {cancel} has no data: the tile comes back blank
maplibregl.addProtocol('custom', (params, callback) => {
  fetch(params.url)
    .then((r) => r.arrayBuffer())
    .then((buffer) => callback(null, buffer))
    .catch((error) => callback(error));
  return { cancel: () => {} };
});

// ✅ v4+ — async (params, abortController) resolving to {data}
maplibregl.addProtocol('custom', async (params, abortController) => {
  const response = await fetch(params.url, { signal: abortController.signal });
  if (!response.ok) throw new Error(`Tile fetch error: ${response.statusText}`);
  const buffer = await response.arrayBuffer();
  return { data: buffer }; // the {data} wrapper is required; a bare buffer is not
});

PMTiles users need no code change. new pmtiles.Protocol().tile is the library's v4 handler wrapped in its v3compat shim (pmtiles ≥ 3.0.0): it checks whether the second argument is an AbortController and adapts, so the single line maplibregl.addProtocol('pmtiles', protocol.tile) is correct on GL JS v3, v4, v5 and v6, and it already resolves to {data: Uint8Array, cacheControl, expires}. A project still on pmtiles 2.x has a callback-only tile and must upgrade the library along with MapLibre. The contract above is what you need when you write your own handler, or when a hand-rolled callback handler broke on the v4 upgrade.

Referencing layers: The style has one source (e.g. sources.tiles) pointing at the .pmtiles URL. Each layer in the layers array that draws from that file uses source: 'tiles' and "source-layer": "layerName", where layerName is the name of a vector layer inside the file (from whatever schema the tiles use). Add multiple style layers with different source-layer values to show roads, labels, etc. from the same file.

Important: The url can be pmtiles://https://... (protocol + HTTPS URL to the .pmtiles file). The library will fetch the file via range requests. Your style must still define glyphs and sprite if you use labels or icons (see maplibre-source-wiring).

Zoom range comes from the header — use url:, not tiles:. A PMTiles archive stores its own min/max zoom in the header. When you reference it with url: 'pmtiles://https://...', the protocol reads that header and hands MapLibre a TileJSON with the correct minzoom/maxzoom, so overzoom past the archive's max works automatically and you never set maxzoom by hand. If you instead hand-wire a tiles: ['pmtiles://.../{z}/{x}/{y}'] template, you bypass that header lookup. The protocol still serves the per-tile requests up to the archive's max — this is not a missing-handler or 404 problem — but MapLibre, given no zoom range, assumes maxzoom: 22 and keeps requesting zoom levels the archive doesn't contain, which come back empty (blank tiles for vector, nothing for raster) instead of overzooming. Always use url:.

❌ A source has exactly two tile-location properties: url and tiles. There is no tileset property in the MapLibre style spec, for vector, raster or raster-dem sources. If a zoom range is wrong, fix which of the two you used; don't reach for a third.

Raster and raster-dem: The same protocol works for raster PMTiles. Use a type: 'raster' source for imagery. For terrain/elevation, use a type: 'raster-dem' source with "encoding": "terrarium" (or "mapbox") so MapLibre can apply hillshade or 3D terrain; then reference it in the style’s terrain property. Example source:

json
"elevation": {
  "type": "raster-dem",
  "url": "pmtiles://https://example.com/elevation.pmtiles",
  "encoding": "terrarium"
}

Using PMTiles with React: Register the protocol once at application startup, not inside each component, so MapLibre has the handler before any map mounts. For example, call maplibregl.addProtocol('pmtiles', protocol.tile) in a root-level effect or when your map provider initializes. On unmount of the last map (or when the app tears down), call maplibregl.removeProtocol('pmtiles') to avoid leaks. See PMTiles for MapLibre GL (Protomaps) for a React-oriented setup.

Hosting PMTiles

Any host that serves the file and supports HTTP Range requests is suitable.

  • AWS S3 — Enable public read (or signed URLs); S3 supports Range. Set Cache-Control and optionally use CloudFront.
  • Cloudflare R2 — S3-compatible; enable public access or use signed URLs. Put behind Cloudflare for caching.
  • GitHub Pages — MapLibre GL JS can load tiles from a .pmtiles file in the same repo as long as the file size is under 100 MB.
  • Netlify / Vercel — Upload the .pmtiles file; static hosting typically supports Range. Check each provider’s file size limits.
  • Any static host — Ensure the server returns Accept-Ranges: bytes and responds correctly to Range headers.

CORS: Browsers will send cross-origin requests to the PMTiles URL. The host must send Access-Control-Allow-Origin: * (or your domain) and Access-Control-Allow-Headers: Range (or allow all). Otherwise MapLibre will fail to load tiles.

Cache headers: For better performance, set long cache for the .pmtiles file (e.g. Cache-Control: public, max-age=31536000 if the file is immutable). CDNs will cache range responses.

The PMTiles CLI

The pmtiles CLI is the official command-line tool for working with PMTiles (and MBTiles for conversion). It’s a single binary with no runtime dependencies—you download it and run it.

Why install and use it:

  • Convert MBTiles to PMTiles — Many tools (tippecanoe, GDAL, martin-cp) output MBTiles. One command turns any .mbtiles file into a .pmtiles file: pmtiles convert in.mbtiles out.pmtiles. This is often the simplest way to get PMTiles when your pipeline already produces MBTiles.
  • Inspect and verify archivespmtiles show <file> prints header and metadata (bounds, zoom range, tile count). pmtiles verify <file> checks archive integrity. Useful for debugging or confirming a file before uploading.
  • Extract subsetspmtiles extract creates a smaller .pmtiles file from an existing one (e.g. by bounding box or zoom range), so you can ship a region or a limited zoom band without regenerating from source.

❌ The subcommand list is closed — don't reach for a plausible-sounding verb. go-pmtiles ships exactly show, tile, verify, extract, merge, serve, convert, cluster, upload, edit and version. There is no info verb and no inspect verb, and nothing sets metadata from the command line except edit. To read an archive's header, bounds and zoom range, the command is pmtiles show; to check integrity, pmtiles verify.

Install: Download the binary for your OS/arch from GitHub Releases (go-pmtiles), or use Docker: protomaps/go-pmtiles.

What it does not do: The CLI only works with tile archives (MBTiles and PMTiles). It does not read GeoJSON, Shapefile, OSM, or other source formats. To create PMTiles from those, use a tool that generates tiles (see Generating PMTiles below) and, if that tool outputs MBTiles, run pmtiles convert to get PMTiles.

Generating PMTiles

Two paths: (1) Convert — The PMTiles CLI converts MBTiles ↔ PMTiles only; it does not read GeoJSON, Shapefile, OSM, or other source formats. (2) Generate from source data — Tools like tippecanoe, Planetiler and ogr2ogr via GDAL read from many file types or databases and produce vector tiles (PMTiles or MBTiles). If they output MBTiles, use pmtiles convert to get PMTiles.

PMTiles CLI (convert only: MBTiles ↔ PMTiles)

See The PMTiles CLI above for why to install it and other commands (show, verify, extract). To convert MBTiles to PMTiles:

bash
pmtiles convert input.mbtiles output.pmtiles

The following tools generate tiles from source data (GeoJSON, OSM, Shapefile, PostGIS, etc.). They output PMTiles or MBTiles; if MBTiles, run pmtiles convert to get PMTiles.

Planetiler (OSM / OpenMapTiles schema)

Planetiler reads OpenStreetMap (or other sources) and outputs PMTiles or MBTiles in the OpenMapTiles schema.

bash
# Example: build a PMTiles file for a region (e.g. from a .osm.pbf download)
java -jar planetiler.jar --area=monaco --output=monaco.pmtiles

See Planetiler docs for area names, custom sources, and schema options. Output is a single .pmtiles file you can upload to S3/R2/static host.

tippecanoe

tippecanoe generates vector tiles from source formats: GeoJSON, FlatGeobuf, CSV. From v2.17 onward it can output PMTiles directly (-o output.pmtiles). You can also output MBTiles and convert with pmtiles convert.

bash
# Direct PMTiles output (v2.17+)
tippecanoe -zg -o output.pmtiles input.geojson
# Or MBTiles then convert: tippecanoe -o output.mbtiles -z 14 input.geojson && pmtiles convert output.mbtiles output.pmtiles

ogr2ogr (GDAL)

GDAL’s ogr2ogr generates tiles from many geospatial formats (Shapefile, PostGIS, GeoJSON, etc.) and can write MBTiles or PMTiles (GDAL 3.8+). Best for smaller datasets; tippecanoe is more efficient for large vector tile sets.

Raster and raster-dem PMTiles

PMTiles supports raster tiles (PNG/JPEG, e.g. satellite or pre-rendered imagery) and raster-dem (elevation/terrain, e.g. Terrarium or Mapbox encoding). Use tools that produce raster or raster-dem PMTiles; the same protocol and hosting apply. In the style use type: 'raster' for imagery or type: 'raster-dem' with "encoding": "terrarium" (or "mapbox") for terrain—see MapLibre Integration above for an example.

Overture Maps

Overture Maps publishes global open map data. Some providers distribute Overture-derived data as PMTiles (e.g. for buildings, places, transportation). You can also build PMTiles from Overture data with Planetiler or other pipelines. Use the PMTiles URL in your MapLibre style as above.

Performance Tips

  • CDN — Serve the .pmtiles file from a CDN (CloudFront, Cloudflare) so range requests are fast globally.
  • Compression — PMTiles stores tiles compressed; the library handles decompression. Ensure the server does not double-compress (e.g. gzip) the whole file in a way that breaks range requests.
  • Multiple files — For very large coverage, split by region into several .pmtiles files and switch the source URL or use multiple sources by bounds.
  • Caching — Set strong cache headers on the file; the browser and CDN will cache range responses.

Related Skills

  • maplibre-tile-sources — Choosing between GeoJSON and tiles for a dataset.
  • maplibre-source-wiring — TileJSON, source-layer, glyphs and sprite, CORS.
  • maplibre-style-patterns — Layer and paint configuration for vector sources (including PMTiles-backed sources). (Not yet in repo.)

References


This skill is a snapshot. Where a primary source contradicts it — the References above, MapLibre's current documentation, or what MapLibre does when you run it — that source wins. Follow it, then report the disagreement, citing the source and your MapLibre version: editing your installed copy helps no one else and is overwritten on the next update.

Frequently asked questions

What does the Maplibre Pmtiles Patterns AI skill do?

Serverless vector and raster tiles with PMTiles for MapLibre GL JS — single-file format, HTTP range requests, hosting on S3/R2/GitHub Pages, generating with Planetiler or tippecanoe, and the pmtiles protocol. Use when you need no tile server or want to host tiles from static storage.

Why use Maplibre Pmtiles Patterns on TypingMind?

Because you install it once and use it with any model. Maplibre Pmtiles Patterns is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Maplibre Pmtiles Patterns in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/maplibre/maplibre-agent-skills/tree/main/skills/maplibre-pmtiles-patterns. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Maplibre Pmtiles Patterns?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Maplibre Pmtiles Patterns?

As many as you like. As long as a model supports skills, you can use Maplibre Pmtiles Patterns with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Maplibre Pmtiles Patterns AI skill free?

It is published on GitHub by maplibre. Check the repository for licensing terms. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇