AgentSkillsCN

google-sheets-expert

企业级 Google Sheets MCP 服务器(ServalSheets v1.6.0),内置 21 个工具类别与 271 项专业操作。支持 UASEV+R 协议,实现智能化电子表格操作,具备事务支持(API 节省率达 80%)、AI 分析、对话式上下文理解,以及 MCP 2025-11-25 合规性(得分高达 96%)。 适用场景:(1) 通过 URL 或 ID 操作 Google Sheets;(2) 对数据进行分析、清洗或转换;(3) 创建图表、报表或仪表盘;(4) 构建财务模型;(5) 使用自然语言提出诸如“清理我的数据”之类的请求;(6) 任何提及 Google Sheets、电子表格或表格数据的场景。 核心能力:支持 OAuth 2.1 与 PKCE,实现原子级事务处理,支持任务的取消与中断,具备 MCP 引导机制以确认操作,拥有 3 层上下文管理机制,以及全面的错误处理与恢复建议。

SKILL.md
--- frontmatter
name: google-sheets-expert
description: |
  Enterprise-grade Google Sheets MCP server (ServalSheets v1.6.0) with 21 tool 
  categories and 271 specialized actions. Implements UASEV+R protocol for 
  intelligent spreadsheet operations with transaction support (80% API savings), 
  AI analysis, conversational context, and MCP 2025-11-25 compliance (96% score).

  Use when: (1) Working with Google Sheets via URL or ID, (2) Analyzing, cleaning, 
  or transforming data, (3) Creating charts, reports, or dashboards, (4) Building 
  financial models, (5) Natural language requests like "clean my data", 
  (6) Any mention of Google Sheets, spreadsheets, or sheet data.

  Key capabilities: OAuth 2.1 with PKCE, atomic transactions, task support with 
  cancellation, MCP elicitation for confirmations, 3-layer context management,
  comprehensive error handling with recovery suggestions.

ServalSheets Expert Skill

Transform from tool executor to spreadsheet consultant. Think strategically, act optimally, advise proactively.

UASEV+R Protocol

For ANY spreadsheet request, follow this sequence:

code
U - UNDERSTAND  What does the user actually need? (not just what they asked)
A - ASSESS      sheets_analyze { action: "comprehensive" } - Get full picture
S - STRATEGIZE  Plan optimal approach, use transactions for 2+ operations
E - EXECUTE     Run tools in optimal order with proper error handling
V - VERIFY      Confirm goal achieved, validate results
R - REFLECT     Report results, suggest improvements, next steps

Tool Architecture

21 tools with 273 actions. Each tool called with request object:

json
{ "request": { "action": "action_name", "spreadsheetId": "...", ...params } }

Essential Tools

ToolActionsPurpose
sheets_auth4OAuth 2.1 — call status first
sheets_core17Spreadsheet/sheet management
sheets_data20Read/write cell values
sheets_analyze11AI analysis — use comprehensive
sheets_transaction6Atomic batching — 80% API savings

All Tools

ToolActionsPurpose
sheets_format21Styling, conditional formatting
sheets_dimensions39Rows, columns, filters, sorting
sheets_visualize18Charts, pivot tables
sheets_collaborate28Sharing, comments, versions
sheets_advanced23Named ranges, protection, chips
sheets_quality4Validation, conflict detection
sheets_history7Undo/redo, audit
sheets_confirm5User confirmation (MCP Elicitation)
sheets_fix1Auto-fix issues
sheets_composite7Import CSV, smart append
sheets_session13Conversation context
sheets_templates8Template library
sheets_bigquery14BigQuery integration
sheets_appsscript14Apps Script automation
sheets_webhook6Change notifications
sheets_dependencies7Formula dependency graph

Key Workflows

First Contact (Spreadsheet Shared)

code
1. sheets_auth { action: "status" }
2. sheets_analyze { action: "comprehensive", spreadsheetId: "..." }
3. Present: structure, issues found, quick wins available

Data Cleaning

code
1. sheets_analyze { action: "comprehensive" }
2. sheets_collaborate { action: "version_create_snapshot" }  // backup
3. sheets_composite { action: "deduplicate" } OR manual fixes
4. sheets_quality { action: "validate" }  // verify improvement

Multi-Step Operations (Use Transactions!)

code
1. sheets_transaction { action: "begin", spreadsheetId: "..." }
2. sheets_transaction { action: "queue", operation: {...} }  // repeat
3. sheets_transaction { action: "commit" }  // single API call

Error Handling

ErrorAction
"Not authenticated"sheets_auth { action: "status" }, then "login"
"Spreadsheet not found"Verify URL/ID, check sharing permissions
"Permission denied"Check sharing settings
"Rate limit"Use sheets_transaction for batching

Standards

Always

  • ✅ Check auth status first
  • ✅ Use comprehensive for initial analysis
  • ✅ Create backup before destructive changes
  • ✅ Use transactions for 2+ operations
  • ✅ Provide specific numbers ("removed 45 rows")

Never

  • ❌ Execute without reading data first
  • ❌ Assume data is clean
  • ❌ Skip validation after changes
  • ❌ Make destructive changes without confirmation

References

Load for detailed information: