W3 MCP LSP Server

W3 MCP LSP Server

Provides JavaScript/TypeScript code intelligence via LSP, enabling goto definition, hover info, and find references through MCP tools.

Category
访问服务器

README

W3 MCP LSP Server

Python MCP server wrapping TypeScript Language Server for JavaScript/TypeScript code intelligence.

Status: ✅ Working with real LSP protocol (not mock)

Features

  • lsp_goto_definition - Jump to symbol definition location
  • lsp_hover - Get type information and documentation
  • lsp_find_references - Find all references/usages of symbol

Supports JavaScript and TypeScript files via TypeScript Language Server.

Quick Start

1. Clean Setup (Important!)

cd /path/to/w3-mcp-server-lsp

# Remove old lockfile and venv
rm -rf uv.lock .venv venv

# Unset old environment variable
unset VIRTUAL_ENV

2. Install Dependencies

# Install TypeScript Language Server
npm install -g typescript typescript-language-server

# Install Python dependencies (using uv)
uv sync

# Install MCP CLI dependencies
uv pip install 'mcp[cli]'

3. Verify Installation

# Check typescript-language-server
which typescript-language-server
typescript-language-server --version

# Check Python env
uv run python -c "from mcp.server.fastmcp import FastMCP; print('✓ MCP ready')"

4. Test with MCP Inspector

# Start MCP Inspector (interactive web UI)
uv run mcp dev server.py

Opens URL like:

http://localhost:6274/?MCP_PROXY_AUTH_TOKEN=...

Features:

  • ✅ Available tools listed in sidebar
  • ✅ Test each tool interactively with JSON input
  • ✅ Real-time request/response viewing
  • ✅ Server logs and debugging
  • ✅ No extra dependencies needed

Usage

Option A: MCP Inspector (Development)

Best way to test and debug:

cd /path/to/w3-mcp-server-lsp

# Start inspector
uv run mcp dev server.py

Opens web UI at http://localhost:5173:

  • See available tools
  • Test each tool with JSON input
  • View request/response in real-time
  • See server logs

Option B: Direct Python

# Run server (stdio mode)
uv run python server.py

Option C: Claude Code Integration

Method 1: From PyPI (Recommended)

Install from PyPI:

pip install w3-mcp-server-lsp
# or
uv pip install w3-mcp-server-lsp

Edit ~/.claude/claude_config.json or ~/.mcp.json:

{
  "mcpServers": {
    "w3-lsp": {
      "type": "stdio",
      "command": "uv",
      "args": ["run", "--with", "w3-mcp-server-lsp", "w3-mcp-server-lsp"],
      "env": {
        "PROJECT_ROOT": "/path/to/your/project"
      }
    }
  }
}

Advantages:

  • ✅ No need to clone the repo
  • ✅ Easy version management
  • ✅ Automatic dependency isolation

Method 2: From Local Source

Edit ~/.claude/claude_config.json:

{
  "mcpServers": {
    "w3-lsp": {
      "type": "stdio",
      "command": "uv",
      "args": ["run", "server.py"],
      "cwd": "/path/to/w3-mcp-server-lsp",
      "env": {
        "PROJECT_ROOT": "/path/to/your/project"
      }
    }
  }
}

Then restart Claude Code.

Tools Documentation

⚠️ IMPORTANT: 0-Indexed Line and Character Numbers

All tools use 0-indexed positioning (start from 0, not 1):

Editor Display Tool Input Formula
Ln 5, Col 10 line: 4 lsp_value = editor_value - 1
Ln 1, Col 1 line: 0 character: 0
Ln 10, Col 5 line: 9 character: 4

Quick Reference:

  • Line 1 in editor = line: 0 in tool
  • Column 1 in editor = character: 0 in tool

How to Calculate Character Position:

Given this code line:

const greeting = "Hello";
0123456789...
  • Position at c in const → character: 0
  • Position at o in const → character: 1
  • Position at g in greeting → character: 6
  • Position at H in "Hello" → character: 18

In VSCode:

  • Click on a character
  • Look at status bar: Ln X, Col Y
  • Use: line: X-1, character: Y-1

lsp_goto_definition

Jump to definition of symbol at specified position.

Input:

{
  "file_path": "fixtures/sample.js",
  "line": 4,
  "character": 9
}

Output:

/path/to/fixtures/sample.js:5:9

lsp_hover

Get type information and documentation for symbol.

Input:

{
  "file_path": "fixtures/sample.js",
  "line": 5,
  "character": 10
}

Output:

(parameter) a: any

lsp_find_references

Find all references/usages of symbol.

Input:

{
  "file_path": "fixtures/sample.js",
  "line": 4,
  "character": 9
}

Output:

Found 2 references:
/path/to/fixtures/sample.js:5:9
/path/to/fixtures/sample.js:14:22

Configuration

PROJECT_ROOT

Specifies the root directory of your project. Files paths in tool calls are relative to this.

Set via:

  1. Environment variable:

    export PROJECT_ROOT="/path/to/your/project"
    uv run python server.py
    
  2. Current directory (default):

    cd /path/to/your/project
    uv run python /path/to/w3-mcp-server-lsp/server.py
    
  3. In .claude/settings.json:

    "env": {
      "PROJECT_ROOT": "/path/to/your/project"
    }
    

Line and Character Numbers (0-indexed)

LSP uses 0-indexed positions:

  • VSCode shows: Ln 5, Col 10
  • LSP needs: line: 4, character: 9

Formula: lsp_value = vscode_value - 1

Project Structure

w3-mcp-server-lsp/
├── server.py           # MCP server entry point
├── lsp_client.py       # LSP protocol implementation
├── pyproject.toml      # Project config
├── fixtures/
│   └── sample.js       # Sample file for testing
├── test_mcp_server.py  # Integration test
├── test_lsp_debug.py   # Debug test with logging
└── README.md

How It Works

Architecture

MCP Client (Claude, IDE, etc.)
    ↓
MCP Server (server.py)
    ↓
LSP Client (lsp_client.py)
    ↓
TypeScript Language Server (subprocess)

LSP Protocol Flow

  1. Initialization

    • Send: initialize request
    • Receive: server capabilities
    • Send: initialized notification
  2. Document Opening

    • Send: textDocument/didOpen notification
    • Content provided inline
  3. Queries

    • Send: textDocument/definition, textDocument/hover, etc.
    • Receive: results from language server
  4. Details

    • Communication: JSON-RPC over stdin/stdout
    • Binary mode: Avoids text encoding issues
    • Notification handling: Skip server notifications, match message IDs

Testing

Interactive Testing (Recommended)

uv run mcp dev server.py

Web UI opens at http://localhost:5173:

  • Test tools visually
  • See real-time results
  • View server logs

Direct Server

uv run python server.py

Runs in stdio mode, ready to connect from Claude Code or other MCP clients.

Troubleshooting

TypeScript Language Server not found

# Install
npm install -g typescript typescript-language-server

# Verify
which typescript-language-server

MCP module not found

# Install dependencies
pip install -e .

# Or manually
pip install mcp pydantic

Server hangs on startup

  • Check if TypeScript Language Server is installed
  • Check terminal for error messages
  • Try: typescript-language-server --stdio directly

Tool returns None or error

  • Verify file path is relative to PROJECT_ROOT
  • Check line/character numbers are 0-indexed
  • Check server logs output in terminal
  • Use MCP Inspector (uv run mcp dev server.py) to see requests/responses

Future Enhancements

  • [ ] Go support (gopls language server)
  • [ ] Python support (pylsp/pyright)
  • [ ] Multiple language servers in single MCP server
  • [ ] Caching of responses
  • [ ] Batch operations

Development

Testing with MCP Inspector

uv run mcp dev server.py

Web UI at http://localhost:5173 shows:

  • Available tools and schemas
  • Real-time request/response
  • Server logs
  • Interactive testing

Running Directly

uv run python server.py

For debugging, check logs output in terminal.

References

License

MIT

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选