你會建立什麼
這篇教學會示範如何把 Cursor 接上 SEO/GEO data，但不把你的 editor 變成不安全的 publishing robot。初學者安全模式是：先用 exports，再建立 normalized files，第三步才是 read-only MCP；只有 approved opportunity report 出來後，才做 page edits。
你會建立：
.cursor/
rules/
seo-geo-data-policy.mdc
seo-geo/
exports/
gsc/
bing/
ga4/
ai-visibility/
snapshots/
reports/
scripts/
normalize-seo-geo-data.mjs
mcp-seo-data-server.mjs
圖說：資料 connector 只提供決策 context，不應自動發布、提交 URL 或修改 production files。
用 data 來輔助 decision。不要讓 data connectors 自動 publish pages、submit URLs、change analytics settings 或 edit production files。
步驟 1：建立 data safety 的 Cursor rule
建立 folders：
mkdir -p .cursor/rules seo-geo/exports/gsc seo-geo/exports/bing seo-geo/exports/ga4 seo-geo/exports/ai-visibility seo-geo/snapshots seo-geo/reports scripts
建立 project rule：
cat > .cursor/rules/seo-geo-data-policy.mdc <<'EOF2'
---
description: SEO/GEO data access and edit safety
globs:
- "seo-geo/**"
- "scripts/**"
- "src/**"
- "app/**"
- "pages/**"
alwaysApply: false
---
For SEO/GEO data tasks:
- read exports from `seo-geo/exports/`
- write normalized snapshots to `seo-geo/snapshots/`
- write reports to `seo-geo/reports/`
- do not invent missing metrics
- do not paste secrets into prompts or files
- do not submit URLs, publish content, deploy, or change production files during data analysis
- separate GSC, Bing, GA4, SERP, and AI visibility signals
- cite source files for every recommendation
- ask before editing website pages
EOF2
如果你的 Cursor version 使用不同的 rules UI，就把同樣 policy 貼進 project rule interface。
步驟 2：加入 sample exports
先從 CSV/JSON files 開始。之後你可以換成 API/MCP。
TODAY=$(date +%F)
cat > seo-geo/exports/gsc/$TODAY-pages.csv <<'EOF2'
url,query,clicks,impressions,ctr,position
/blog/reporting-automation,reporting automation software,34,4200,0.008,8.2
/features/reporting,reporting tool for startups,18,1900,0.009,11.4
EOF2
cat > seo-geo/exports/bing/$TODAY-queries.csv <<'EOF2'
query,clicks,impressions,ctr,position
reporting automation software,4,230,0.017,6.8
automate weekly reports,2,180,0.011,9.3
EOF2
cat > seo-geo/exports/ai-visibility/$TODAY-prompts.json <<'EOF2'
[
{
"prompt": "What is the best reporting automation software for startups?",
"surface": "manual-check",
"brand_mentioned": false,
"citation_url": null,
"competitors": ["Competitor A", "Competitor B"]
}
]
EOF2
步驟 3：先 normalize data，再請 Cursor 做 strategy
建立 normalizer：
cat > scripts/normalize-seo-geo-data.mjs <<'EOF2'
#!/usr/bin/env node
import { readFileSync, writeFileSync, mkdirSync, readdirSync, existsSync } from 'node:fs';
import { join } from 'node:path';
const date = process.argv[2] || new Date().toISOString().slice(0, 10);
mkdirSync('seo-geo/snapshots', { recursive: true });
const rows = ['source,date,query_or_prompt,target_url,metric,value,evidence_url,confidence,raw_file'];
function latestFile(dir, suffix) {
if (!existsSync(dir)) return null;
return readdirSync(dir).filter(f => f.includes(date) && f.endsWith(suffix))[0] || null;
}
function addCsv(source, dir, suffix, metricField) {
const file = latestFile(dir, suffix);
if (!file) return;
const raw = readFileSync(join(dir, file), 'utf8').trim().split(/
?
/);
const headers = raw.shift().split(',');
for (const line of raw) {
const values = line.split(',');
const item = Object.fromEntries(headers.map((h, i) => [h, values[i] || '']));
rows.push([source, date, item.query || '', item.url || '', metricField, item[metricField] || '', item.url || '', 'medium', file].map(JSON.stringify).join(','));
}
}
addCsv('gsc', 'seo-geo/exports/gsc', '.csv', 'clicks');
addCsv('bing', 'seo-geo/exports/bing', '.csv', 'clicks');
const aiFile = latestFile('seo-geo/exports/ai-visibility', '.json');
if (aiFile) {
const items = JSON.parse(readFileSync(join('seo-geo/exports/ai-visibility', aiFile), 'utf8'));
for (const item of items) {
rows.push(['ai-visibility', date, item.prompt || '', '', 'brand_mentioned', String(Boolean(item.brand_mentioned)), item.citation_url || '', 'low', aiFile].map(JSON.stringify).join(','));
}
}
const out = `seo-geo/snapshots/${date}-normalized.csv`;
writeFileSync(out, rows.join('
'));
console.log(`Wrote ${out}`);
EOF2
chmod +x scripts/normalize-seo-geo-data.mjs
node scripts/normalize-seo-geo-data.mjs $(date +%F)
步驟 4：要求 Cursor 產生 opportunity report，不要直接 edits
在 Cursor Agent 貼上：
Use the Cursor rule `seo-geo-data-policy`.
Read the latest normalized CSV in `seo-geo/snapshots/`.
Create `seo-geo/reports/YYYY-MM-DD-opportunity-report.md`.
For each recommendation include:
- source file
- URL or prompt
- signal
- interpretation
- confidence
- action type: refresh, create, internal link, technical ticket, monitor, ignore
- files likely involved
- approval needed
Do not edit website pages yet. Do not publish. Do not submit URLs.
好的 output 會引用 files。弱 output 只會給 generic「improve SEO」建議。如果太弱，要求 Cursor 重做並引用每一列 source row。
步驟 5：file workflow 穩定後，再加 placeholder MCP server
建立一個 placeholder，先明確定義 permissions：
cat > scripts/mcp-seo-data-server.mjs <<'EOF2'
#!/usr/bin/env node
// Placeholder for a read-only SEO/GEO MCP server.
// Future tools: get_gsc_pages, get_bing_queries, get_ga4_landing_pages,
// get_ai_visibility_snapshot. No write, submit, publish, or delete tools.
console.error('SEO/GEO MCP placeholder: read-only tools only.');
setInterval(() => {}, 1000);
EOF2
chmod +x scripts/mcp-seo-data-server.mjs
接著問 Cursor：
Design a read-only MCP configuration for this project using `scripts/mcp-seo-data-server.mjs`.
Before editing settings, explain:
- where Cursor expects MCP configuration in this environment
- tool names
- input and output schemas
- credential handling
- how to disable write actions
- how to test with one read-only query
Do not add API credentials. Do not create write tools.
最後 setup 請使用 Cursor 目前的 MCP settings UI 或 config format，因為 Cursor 的 MCP configuration surface 可能因版本與 workspace 而不同。
步驟 6：報告完成後，只核准一個 page edit
Opportunity report 建好後，只核准一頁：
From `seo-geo/reports/YYYY-MM-DD-opportunity-report.md`, approve only the recommendation for `/features/reporting`.
Allowed action: metadata and opening answer block refresh.
Allowed files: [list files]
Not approved: schema, canonical, robots, sitemap, redirects, analytics, shared template edits.
Use data as private decision context. Do not cite analytics numbers in public copy unless explicitly approved.
Run validation and summarize changed files.
這樣能把 data analysis 和 publishing 分開。
Troubleshooting
| Problem | Likely cause | Fix |
|---|---|---|
| Cursor 捏造 missing metrics | Rule 沒載入 | 明確提到 rule，並要求 source files |
| Opportunity report 太 vague | 沒有 normalized snapshot | 先跑 normalizer |
| MCP setup 要求 broad tools | Tool boundary 不清楚 | 只允許 read-only tools |
| Public copy 暴露 analytics | Data 被當成公開 proof | 在 prompt 加上 private decision context |
| Cursor edit 太多 files | Approval 太寬 | 只核准一個 URL 和 exact files |
FAQ
一開始需要 MCP 嗎？
不需要。CSV exports 就夠了。當 workflow 重複、data freshness 變重要時，MCP 才有價值。
Cursor 應該 submit URLs 到 Bing 或 update CMS content 嗎？
第一版不要。保持 read-only actions，直到團隊信任 workflow。
Cursor 可以分析 AI visibility snapshots 嗎？
可以。用 JSON 或 CSV 保存 prompt、surface、answer summary、brand mention、citation URL、competitors 和 date。
AI coding agents for SEO/GEO 學習路徑
Cursor track：
- Cursor vs Windsurf vs Gemini CLI：SEO/GEO 自動化怎麼選
- 如何設定 Cursor Rules 做 SEO/GEO 工作
- 如何用 Cursor Plan Mode 安全更新 SEO 頁面
- 如何用 MCP 將 Cursor 接上 SEO 資料
- 如何用 Cursor Cloud Agents 做 SEO/GEO PR Reviews
- Cursor、Windsurf、Gemini CLI、Codex、Claude Code 的 SEO/GEO 營運模型
Sources and notes
- Cursor documentation hub: Cursor Docs ，包含 Agent、Rules、MCP、Skills、CLI 與 background/agent capabilities 的公開文件。
- Related Auspia guide: Cursor vs Windsurf vs Gemini CLI for SEO/GEO Automation 。
Author: Leo Harrington, Auspia SEO Analytics Translator for 500+ Executive Reports。Leo 撰寫如何把 search 與 AI visibility data 轉成團隊可執行 decisions。
