skill
qgis-map-workflow
Turn a spreadsheet into a publication-grade, editable QGIS map by chatting with your AI — via the QGIS MCP plugin. 用 AI 对话生成可编辑的出版级 QGIS 地图。
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/maydengximin-sketch/qgis-map-workflow itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 21, 2026qgis-map-workflow
Turn a spreadsheet into a publication-grade, editable QGIS map — by chatting with your AI. 把一张表格,对着 AI 说句话,变成出版级、可编辑的 QGIS 地图。
English
An LLM skill that drives QGIS through the QGIS MCP plugin to make
publication-grade maps from any LLM client (Claude Code / Claude Desktop / Cursor /
Codex). You bring tabular or spatial data and say what you want; the skill detects the
MCP connection, imports your data, checks the coordinates, picks the right map type,
styles it against a reproducible cartographic standard, self-checks the result by
screenshot, and hands back a live, editable .qgz project plus an export-ready
print layout or atlas — no manual QGIS clicking required.
Originally built for a children's urban-emotion mapping project; generalized into a
broadly reusable QGIS map-making workflow. That original use case ships as a worked
example (§13 of SKILL.md).
Note on language:
SKILL.mdand two of the threereferences/files are written in Chinese (the skill produces figure text in the user's language and defaults to Chinese for Chinese users).references/algorithm-reference.mdstays in English to keep algorithm IDs / class names exact.
What it can make
QGIS's main map families, via a Map Type Router (§3 of SKILL.md):
- Vector symbology — categorized, graduated/choropleth, proportional-symbol, dot-density, rule-based, per-feature chart overlays
- Density & hotspot — KDE heatmaps, hexbin aggregation, Getis-Ord Gi* hotspots
- Interpolation surfaces — IDW, TIN, kriging, contours
- Raster & terrain — pseudocolor, hillshade/slope/aspect, RGB composites
- Proximity & geometry — buffers, Voronoi/Thiessen, isochrones
- Flow & network — OD flow lines, routing
- Time — temporal animations, time small-multiples
- Composition — labeled/annotated maps, bivariate choropleth, print layouts, atlases
Each common type comes with an explicit design spec (classification, color ramp,
opacity, interpolation parameters, clipping, DPI) so results are consistent and
reproducible. It runs autonomously: drives QGIS through the MCP plugin's real tools,
screenshots its own output to self-check before exporting, delivers a live, editable
.qgz project (not a flattened image), and for multi-map jobs proposes a few style
options to choose from, then batch-produces the whole set.
Requirements
- QGIS (LTR recommended) — https://qgis.org/download/
- QGIS MCP plugin — install from QGIS Plugin Manager, then
Start Server(port 9876) - An LLM client with this skill installed and the QGIS MCP server connected
uv/uvxfor the MCP server — https://docs.astral.sh/uv/
See SKILL.md §1 for per-client (Claude Code / Desktop / Cursor / Codex) setup.
Quick start
- Install QGIS + the QGIS MCP plugin; click
Start Server. - Connect your client to the MCP server (
SKILL.md§1). - Try it on the bundled synthetic data:
Use the qgis-map-workflow skill. Inspect examples/sample_points.csv, confirm
lon/lat, and list which map types I can make. Then, with QGIS open and the MCP
server started, make a graduated choropleth on `value` and a KDE heatmap.
No MCP yet? The model can still produce a static preview sketch for discussion — but the QGIS output is the real deliverable, never the sketch.
Repository layout
SKILL.md 精简操作手册 / Lean operating manual (start gate, router, MCP exec, QA, editable output, batch)
README.md This file
LICENSE MIT
CHANGELOG.md Version history
references/
map-specs.md 各图型设计规范、选择/标注规则、排错、范例(中文)
algorithm-reference.md MCP tool list + Processing algorithm IDs + PyQGIS renderers (English)
cartographic-quality.md 制图美观+合规标准,QA 自检依据(中文)
examples/
sample_points.csv Synthetic point dataset to try cold
areas.geojson Synthetic polygons for a choropleth demo
README.md Column reference + example prompts
diagrams/
workflow.md Workflow diagram
assets/
poster*.png / *.html 小红书图文海报 / promo posters (not required to run the skill)
docs/
README.md
Safety
The skill never edits your source data, won't touch QGIS unless you ask, and reports
exactly what it changed and where it saved (SKILL.md §11). Keep real field data out
of any public fork — only the synthetic examples/ data is meant to be shared.
License
MIT © Zoey — see LICENSE.
中文
这是一个 LLM 技能(skill),通过 QGIS MCP 插件驱动 QGIS,让你在任意 LLM
客户端(Claude Code / Claude Desktop / Cursor / Codex)里制作出版级地图。你提供表格或
空间数据、说出想要的效果;技能会自动检测 MCP 连接、导入数据、校验坐标、选择合适的图型、
按可复现的制图标准上色,自己截图自检,最后交还一个实时可编辑的 .qgz 工程,外加
可直接导出的打印布局或地图集——全程无需手动点 QGIS。
最初为一个"儿童城市情绪地图"项目而建,后泛化成可广泛复用的 QGIS 制图工作流;原始用例作为
完整范例保留在 SKILL.md 的 §13。
关于语言:
SKILL.md与references/中两个文件为中文(技能用用户的语言生成图面 文字,中文用户默认中文)。references/algorithm-reference.md保留英文,以保证算法 ID / 类名精确无误。
能做哪些图
通过 SKILL.md §3 的图型路由表覆盖 QGIS 主要图型:
- 矢量符号化 —— 分类图、分级/等值区域图、比例符号、点密度、基于规则、每要素图表叠加
- 密度与热点 —— KDE 核密度热力图、六边形聚合、Getis-Ord Gi* 热点
- 插值面 —— IDW、TIN、克里金、等值线
- 栅格与地形 —— 伪彩色、山体阴影/坡度/坡向、RGB 合成
- 邻近与几何 —— 缓冲区、泰森多边形、等时圈
- 流向与网络 —— OD 流线、路径分析
- 时间 —— 时间动画、时间小多图
- 排版组合 —— 标注地图、双变量等值区域图、打印布局、地图集
每个常用图型都配有明确的设计规范(分级方法、色带、不透明度、插值参数、裁剪、DPI),
保证多次运行、多人协作结果一致、可复现。技能自主运行:用 MCP 插件的真实工具驱动 QGIS、
导出前自己截图自检、交付可编辑的 .qgz 工程(而非压平图片);多图任务会先给几套
风格供你挑,再批量出整组图。
环境要求
- QGIS(推荐 LTR 长期支持版)—— https://qgis.org/download/
- QGIS MCP 插件 —— 在 QGIS 插件管理器安装,点
Start Server(端口 9876) - 一个 LLM 客户端,装好本技能并连上 QGIS MCP 服务器
- MCP 服务器需要
uv/uvx—— https://docs.astral.sh/uv/
各客户端(Claude Code / Desktop / Cursor / Codex)的配置见 SKILL.md §1。
快速开始
- 安装 QGIS + QGIS MCP 插件,点
Start Server。 - 把客户端连到 MCP 服务器(
SKILL.md §1)。 - 用自带的合成数据试跑:
使用 qgis-map-workflow 技能。检查 examples/sample_points.csv,确认 lon/lat,
列出我能做哪些图型。然后在 QGIS 已打开、MCP 服务器已启动的情况下,按 `value`
做一张分级等值区域图,再做一张 KDE 热力图。
还没装 MCP?模型仍可生成一张静态预览草图供讨论——但 QGIS 成品才是真正的交付物, 草图绝不是。
安全
技能绝不修改你的源数据、不经请求不碰 QGIS,并会报告它改了什么、存到了哪里(SKILL.md §11)。
公开 fork 请勿放入真实字段数据——只有合成的 examples/ 数据是用来分享的。
许可
MIT © Zoey —— 见 LICENSE。