MapMap Agent Skills
Skills MCP server Licence: MIT
Knowledge modules that teach AI coding agents how to build with MapMap — the self-hostable navigation platform for fleets and AI agents.
Skills are the how-to knowledge layer. They complement the
MapMap MCP server, which gives agents live
tools (routing, ADR compliance, geocoding, matrix, optimisation and map
styling at https://mcp.mapmap.ai/mcp). Install both: the MCP server lets
your agent call MapMap; these skills teach it to build with MapMap correctly
on the first try.
Install
# All skills
npx skills add Mapmapai/mapmap-agent-skills
# One skill
npx skills add Mapmapai/mapmap-agent-skills --skill mapmap-truck-adr-routing
# See what's available
npx skills add Mapmapai/mapmap-agent-skills --list
Works with Claude Code, Cursor, Codex, VS Code with Copilot and other agents
that read SKILL.md modules. Manual install for Claude Code:
git clone https://github.com/Mapmapai/mapmap-agent-skills.git
cd your-project
mkdir -p .claude && ln -s ../mapmap-agent-skills/skills .claude/skills
Skills
| Skill | Teaches |
|---|---|
mapmap-mcp-setup | Connecting any MCP client to MapMap — hosted endpoint, self-host, per-client config, tool reference |
mapmap-truck-adr-routing | Truck routing with dimensional limits and ADR dangerous-goods tunnel codes — parameters, semantics, gotchas |
mapmap-web-maps-integration | @mapmap/maps in web apps — maps, routing, turn-by-turn guidance, navigation camera, Studio themes |
mapmap-map-design | Designing branded map styles — brand → 19-slot palette ("make maps like airbnb.com"), legibility rules, publishing immutable styles |
mapmap-fleet-optimisation | Multi-vehicle VRP — vehicles, jobs, shipments, time windows, capacities, truck/ADR constraints in the matrix, the 200-location cap |
mapmap-fleet-sync | The morning-dispatch loop against a telematics platform you already run — read today's stops and vehicles, geocode and sanity-check, optimise, write the ordered route back or export it, then follow it live with the stateless progress endpoint |
mapmap-offline-territories | Offline maps — signed territory packages, verifying-key pinning, differential OTA updates, the download allowance |
mapmap-migrate-from-mapbox | Moving a Mapbox GL / Directions API app to MapMap — endpoint mapping, tokens to keys, style migration |
mapmap-migrate-from-google-maps | Moving a Google Maps Platform app to MapMap — Routes/Matrix/Geocoding mapping, what ports and what has no replacement |
mapmap-x402-payments | How agents pay per call — 402 vs 429 dispatch, prepaid credit, inline x402 where configured, refund and polling rules |
mapmap-self-host-ops | Running the whole stack yourself — Docker Compose distro, keys, territories, production notes |
Claude Code plugin
This repository is also an installable Claude Code plugin (named mapmap)
and its own plugin marketplace:
claude plugin marketplace add Mapmapai/mapmap-agent-skills
claude plugin install mapmap@mapmap
Installing the plugin gives you:
- All 11 skills, namespaced as
/mapmap:<skill-name>(e.g./mapmap:mapmap-truck-adr-routing). - The hosted MapMap MCP server (
https://mcp.mapmap.ai/mcp) via the bundled.mcp.json— live routing, ADR, matrix, optimisation and map styling tools. - An opt-in retro reminder hook — disabled by default. A small
Stophook that does nothing unless you setMAPMAP_RETRO_HOOK=1. When enabled, and only when a session mentions MapMap, it reminds the agent (once per session) to offer you an integration retro. It never submits anything and makes no network calls — seehooks/README.md.
Feedback programme
Each skill ends with an optional, ask-first final step: if you approve,
your agent can send MapMap a structured integration retro — what it built,
problems hit, gotchas, wins, and docs gaps — via the
submit_integration_retro MCP tool or POST https://api.mapmap.ai/v1/feedback (API-key auth, limited to 5 per key per
day). This sends only those structured fields — never your
conversation, code, or credentials — and your agent should ask you before
sending (or skip it). It exists so real integration friction reaches the
people who can fix it. Full programme details, fields, and retention:
mapmap.ai/legal/agent-feedback.
The machine surface
MapMap is built agent-first. Before scraping anything, an agent should read:
https://api.mapmap.ai/llms.txt— orientation: endpoints, auth, pricing, MCPhttps://mapmap.ai/pricing.json— machine-readable priceshttps://api.mapmap.ai/openapi.json— the authoritative API contract- Every docs page is served as raw markdown: append
.mdto the URL
Keys are self-serve (POST /v1/keys, one call, card-free) and payments are
machine-native (x402 on 402). See mapmap.ai/agents.
Licence
MIT. Map data referenced in examples derives from OpenStreetMap — anything you render or republish must credit "© OpenStreetMap contributors".