MCP Server for VS Code
Enables AI assistants to interact with VS Code for language intelligence, debugging, and code execution.
README
MCP Server for VS Code
A VS Code extension that provides a Model Context Protocol (MCP) server, enabling AI assistants to interact with your VS Code environment for language intelligence, debugging, and code execution.
Features
-
Language Intelligence: Access VS Code's language server features including:
- Go to definition
- Find references
- Diagnostics (errors and warnings)
- Symbol search
- Call hierarchy
-
Debugging Support: Control VS Code's debugger programmatically:
- Start/stop debug sessions
- Set and manage breakpoints
- Step through code (into/over/out)
- Inspect variables and call stacks
- Evaluate expressions in debug context
Installation
Alpha Testing
Step 1: Install VS Code Extension
Download the .vsix file from Releases and install:
- In VS Code: Extensions →
...menu → Install from VSIX - Or via command line:
code --install-extension mcp-server-vscode-*.vsix
Step 2: Configure Claude Desktop
The MCP server runs directly from GitHub using npx. Add this to your Claude config:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"vscode": {
"command": "npx",
"args": ["github:malvex/mcp-server-vscode"]
}
}
}
Step 3: Restart Claude Desktop
That's it! The VS Code tools are now available in Claude.
Configure Claude Code (CLI)
For Claude Code users, run this one-liner:
claude mcp add-json vscode '{"type":"stdio","command":"npx","args":["github:malvex/mcp-server-vscode"]}' -s user
Usage
Once installed, the extension shows the MCP server status in the VS Code status bar (bottom right).
To start the MCP server: Click on "VS Code MCP: Stopped" in the status bar. It will change to "VS Code MCP: 8991" when running.
The status bar indicates:
- VS Code MCP: 8991 - Server is running on port 8991
- VS Code MCP: Stopped - Server is not running
Click the status bar item to toggle the server on/off.
How It Works
┌─────────────┐ stdio ┌──────────────────┐ HTTP ┌─────────────┐
│ Claude │ ◄────────────► │ MCP Standalone │ ◄───────────► │ VS Code │
│ Desktop │ │ Server │ :8991 │ Extension │
└─────────────┘ └──────────────────┘ └─────────────┘
- VS Code Extension provides an HTTP API on port 8991
- MCP Standalone Server acts as a bridge, converting stdio ↔ HTTP
- Claude Desktop communicates with the standalone server via stdio
Troubleshooting
If Claude can't connect to VS Code:
- Check VS Code is running with the extension active
- Check the status bar shows "VS Code MCP: 8991"
- Test the MCP server: Run
npx github:malvex/mcp-server-vscodein terminal - Check firewall isn't blocking localhost:8991
- Try manually starting the MCP server in VS Code (Cmd/Ctrl+Shift+P → "Start MCP Server")
Available Tools
The extension provides 25 tools organized into three main categories:
Language Intelligence Tools (7 tools)
| Tool | Description | Main Parameters | Example |
|---|---|---|---|
| hover | Get hover information (type info, documentation) for a symbol by name | symbol (required), uri (optional), format (optional) |
hover({ symbol: "calculateSum" }) |
| definition | Find where a symbol is defined. Instantly jumps to declarations | symbol (required), format (optional) |
definition({ symbol: "Calculator" }) |
| references | Find all references to a symbol. Superior to grep - finds semantic references | symbol (required), includeDeclaration (optional), format (optional) |
references({ symbol: "process" }) |
| callHierarchy | Analyze what calls a function or what a function calls | symbol (required), direction (required: 'incoming'|'outgoing'|'both'), uri (optional), format (optional) |
callHierarchy({ symbol: "initialize", direction: "incoming" }) |
| symbolSearch | Search for symbols (classes, functions, variables) across the workspace | query (required), kind (optional), format (optional) |
symbolSearch({ query: "Controller", kind: "class" }) |
| workspaceSymbols | Get a complete map of all symbols in the workspace | includeDetails (optional), filePattern (optional), maxFiles (optional), format (optional) |
workspaceSymbols({ filePattern: "**/*.ts" }) |
| diagnostics | Get all errors and warnings for a file or workspace | uri (optional), format (optional) |
diagnostics({}) |
Refactoring Tools (1 tool)
| Tool | Description | Main Parameters | Example |
|---|---|---|---|
| refactor_rename | Rename a symbol across all files. Automatically updates all references and imports | symbol (required), newName (required), uri (optional), format (optional) |
refactor_rename({ symbol: "OldName", newName: "NewName" }) |
Debug Tools (17 tools)
Breakpoint Management
| Tool | Description | Main Parameters | Example |
|---|---|---|---|
| debug_setBreakpoint | Set breakpoints by symbol name or file/line with optional conditions | symbol OR (file AND line), condition (optional), hitCondition (optional), logMessage (optional), format (optional) |
debug_setBreakpoint({ symbol: "processData", condition: "items.length > 100" }) |
| debug_toggleBreakpoint | Toggle a breakpoint on/off at a specific location | symbol OR (file AND line), format (optional) |
debug_toggleBreakpoint({ file: "app.js", line: 25 }) |
| debug_listBreakpoints | List all breakpoints in the workspace | format (optional) |
debug_listBreakpoints({}) |
| debug_clearBreakpoints | Clear all breakpoints from the workspace | format (optional) |
debug_clearBreakpoints({}) |
Session Management
| Tool | Description | Main Parameters | Example |
|---|---|---|---|
| debug_status | Get current debug session status and active threads | format (optional) |
debug_status({}) |
| debug_listConfigurations | List available debug configurations from launch.json | format (optional) |
debug_listConfigurations({}) |
| debug_startSession | Start a debug session using a configuration | configuration (optional), format (optional) |
debug_startSession({ configuration: "Launch Program" }) |
| debug_stopSession | Stop the active debug session | format (optional) |
debug_stopSession({}) |
Runtime Control
| Tool | Description | Main Parameters | Example |
|---|---|---|---|
| debug_pauseExecution | Pause the running program | threadId (optional), format (optional) |
debug_pauseExecution({}) |
| debug_continueExecution | Continue execution from current breakpoint | threadId (optional), allThreads (optional), format (optional) |
debug_continueExecution({}) |
| debug_stepOver | Step over the current line of code | threadId (optional), format (optional) |
debug_stepOver({}) |
| debug_stepInto | Step into the function call at current line | threadId (optional), format (optional) |
debug_stepInto({}) |
| debug_stepOut | Step out of the current function | threadId (optional), format (optional) |
debug_stepOut({}) |
Inspection and Evaluation
| Tool | Description | Main Parameters | Example |
|---|---|---|---|
| debug_getCallStack | Get the current call stack with source locations | threadId (optional), startFrame (optional), levels (optional), format (optional) |
debug_getCallStack({ levels: 10 }) |
| debug_inspectVariables | Inspect variables in the current scope during debugging | threadId (optional), frameId (optional), scope (optional: 'all'|'locals'|'globals'|'closure'), filter (optional), format (optional) |
debug_inspectVariables({ scope: "locals" }) |
| debug_evaluateExpression | Evaluate an expression in the debug context | expression (required), frameId (optional), context (optional), format (optional) |
debug_evaluateExpression({ expression: "user.permissions" }) |
| debug_getOutput | Get debug console output | category (optional), filter (optional), limit (optional), format (optional) |
debug_getOutput({}) |
Tool Features
All tools support:
- Compact format - Optimized for AI token efficiency
- Detailed format - Full data for complex analysis
- Symbol-based navigation - Work with names instead of file/line numbers
- Workspace-wide operations - Not limited to single files
- Language server integration - Accurate semantic understanding
Usage Examples for AI Assistants
When connected via MCP, AI assistants can use these tools to help users with development tasks:
Finding and Understanding Code
User: "What does the handleRequest function do?"
AI uses: hover({ symbol: "handleRequest" })
→ Gets type signature and documentation without reading entire files
User: "Where is the DatabaseConnection class defined?"
AI uses: definition({ symbol: "DatabaseConnection" })
→ Instantly finds the file and line where it's declared
User: "Show me all places where processPayment is called"
AI uses: callHierarchy({ symbol: "processPayment", direction: "incoming" })
→ Gets complete list of callers with their locations
Refactoring
User: "Rename the oldMethodName method to newMethodName everywhere"
AI uses: refactor_rename({ symbol: "oldMethodName", newName: "newMethodName" })
→ Safely renames across all files, updating imports and references
Debugging
User: "Help me debug why the server crashes"
AI uses: debug_listConfigurations({})
→ Shows available debug configurations
AI uses: debug_startSession({ configuration: "Debug Server" })
→ Starts the debug session
User: "Set a breakpoint where errors are handled"
AI uses: debug_setBreakpoint({ symbol: "handleError" })
→ Sets breakpoint on the function
User: "What's the value of the user object here?"
AI uses: debug_inspectVariables({ scope: "locals", filter: "user" })
→ Shows current value of user variable in debug context
User: "Why is this condition true?"
AI uses: debug_evaluateExpression({ expression: "users.length > 0 && isActive" })
→ Evaluates the expression in current debug scope
Development
Building from Source
# Clone the repository
git clone https://github.com/malvex/mcp-server-vscode.git
cd mcp-server-vscode
# Install dependencies
npm install
# Build everything
npm run compile
npm run package
# Package the VS Code extension
npx vsce package
Testing Local Changes
To test your local development version:
- VS Code Extension: Press F5 in VS Code to launch Extension Development Host
- MCP Server: Update Claude config to use local path:
{
"mcpServers": {
"vscode": {
"command": "node",
"args": ["/path/to/mcp-server-vscode/out/mcp/standalone-server.js"]
}
}
}
License
This project is licensed under the MIT License - see the LICENSE file for details.
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。