Seta MCP
Enables Salesforce developers to create code and configuration using local documentation, with optional semantic search.
README
Seta MCP - Up-to-date Local Docs for Coding
Assist Salesforce developers to create code and configuration! This is a MCP server that you can run locally or host remotely.
Designed to fetch documentation from your local file system, giving you complete control over the context provided to your AI assistant.
Inspired by context7 - but this application uses local documentation library.
Why use this?
Seta MCP server consists of three primary components:
- MCP server code (this repository)
- Open / public, internal or proprietary documentation for Apex, LWC, configuration etc.
- NEW: Semantic search indexer for intelligent document discovery
Seta MCP helps you to -
- Work entirely offline.
- Have fine-grained control over the exact documentation versions and content.
- Quickly iterate on documentation and have it immediately available to your LLM.
- NEW: Find relevant documentation using semantic search, even when exact keywords don't match.
💡 How It Works?
This MCP server reads documentation from a local directory structure you define.
- Set
LOCAL_DOCS_PATH: You specify a root directory for your documentation via theLOCAL_DOCS_PATHenvironment variable. - Organize Libraries: Inside
LOCAL_DOCS_PATH, each library or documentation set should reside in its own subdirectory. - Enhanced
manifest.json: Each library subdirectory must contain amanifest.jsonfile with rich metadata including difficulty levels, use cases, learning paths, and semantic groups. - Optional Semantic Indexing: Run
seta-indexerto create a vector database for semantic search capabilities.
When you prompt your LLM (e.g., in VSCode) with use latest Apex docs, use seta or just reference any of the below tools in the prompt:
- The
get-library-idtool searches themanifest.jsonfiles in yourLOCAL_DOCS_PATHto find matching libraries. - The
get-library-docstool then fetches the content of the specified document file from your local file system.
🛠️ Getting Started
Requirements
- Node.js >= v18.0.0
- VSCode, Cursor, Windsurf, Claude Desktop, or another MCP Client.
- A local directory containing your documentation, structured as described below.
- The
LOCAL_DOCS_PATHenvironment variable must be set to point to this directory.
Local Documentation Structure
Your LOCAL_DOCS_PATH should point to a root directory. Inside this, each library has its own folder:
<your_LOCAL_DOCS_PATH>/
├── my-awesome-lib/
│ ├── manifest.json
│ ├── main_guide.md
│ ├── .seta_lancedb/ # Created by seta-indexer (optional)
│ └── topics/
│ └── report.md
│ └── advanced_reports.txt
└── another-tool/
├── manifest.json
└── quick_start.md
Enhanced manifest.json Format
Each library directory (e.g., my-awesome-lib/) must contain a manifest.json file in the following enhanced format:
{
"name": "My Awesome Library",
"description": "A brief description of what this library does.",
"version": "2.1.0",
"default_doc": "main_guide.md", // File to use if no topic is specified
"topics": [
{
"name": "report",
"file": "report.md",
"tags": ["report"],
"related": ["advanced"],
"prerequisites": [],
"leads_to": ["advanced"],
"difficulty": "beginner",
"use_cases": ["basic reporting", "data visualization"],
"code_patterns": ["SOQL queries", "report types"]
},
{
"name": "advanced",
"file": "topics/advanced_reports.txt",
"tags": ["advanced"],
"related": ["report"],
"prerequisites": ["report"],
"leads_to": [],
"difficulty": "advanced",
"use_cases": ["complex analytics", "custom report types"],
"code_patterns": ["dynamic SOQL", "custom report builders"]
}
],
"semantic_groups": {
"fundamentals": ["report"],
"advanced_features": ["advanced"]
},
"learning_paths": {
"beginner": ["report"],
"expert": ["report", "advanced"]
},
"totalSnippets": 25 // Optional: for display purposes
}
Enhanced topic properties:
prerequisites(array of strings, optional): Topics that should be understood firstleads_to(array of strings, optional): Topics that naturally follow this onedifficulty(string, optional): "beginner", "intermediate", or "advanced"use_cases(array of strings, optional): Practical applications of this topiccode_patterns(array of strings, optional): Code patterns or techniques covered
New library-level properties:
semantic_groups(object, optional): Named collections of related topicslearning_paths(object, optional): Suggested sequences of topics for different skill levels
How enhanced topic lookup works:
- When searching, the server matches your query against library name, description, topic names, tags, use_cases, and code_patterns.
- When fetching documentation for a topic, the server automatically includes context from prerequisites, related topics, and next steps.
- You can request entire semantic groups or learning paths by name.
🔍 Semantic Search with seta-indexer
Development
Clone the project and install dependencies:
npm i
Build (assuming main.ts is in the root and tsconfig.json is configured for outDir: "./dist" and rootDir: "./"):
npm run build
This will compile main.ts (and files in lib/ and indexer/) to dist/main.js and dist/indexer/cli.js.
Local Configuration Example (for development)
To run your local development version with an MCP client like Cursor:
{
"mcpServers": {
"seta_dev": {
"command": "npx",
"args": ["tsx", "/path/to/your/project/dist/main.js"],
"env": {
"LOCAL_DOCS_PATH": "/absolute/path/to/your/test_documentation_root", // Essential for testing
"DEFAULT_MAX_TOKENS": "5000" // Optional
}
}
}
}
Testing with MCP Inspector
Ensure LOCAL_DOCS_PATH is set in your environment.
# For published version
# export LOCAL_DOCS_PATH="/path/to/your/docs"
npx -y @modelcontextprotocol/inspector npx seta-mcp@latest
Disclaimer
This project relies on the documentation you provide. The quality, accuracy, and completeness of the generated context are directly dependent on your local documentation files. Use at your own discretion and risk.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。