vscode-mcp
Exposes VS Code terminals as MCP tools, allowing AI agents to create, execute commands, read output, list, and kill terminals across multiple workspaces.
README
VS Code MCP Extension
Exposes VS Code terminals as MCP tools for AI agents.
Architecture
┌──────────────────────────────────────────────────────┐
│ Central Server (Python, localhost:9876) │
│ - Single MCP endpoint: /mcp │
│ - Routes by workspace name │
│ - Manages extension registry │
└──────────────────────────────────────────────────────┘
↑ HTTP ↑
│ │
┌────────┴──────┐ ┌────────┴──────────┐
│ VS Code #1 │ │ VS Code #2 │
│ Extension │ │ Extension │
│ Workspace: │ │ Workspace: │
│ "project-a" │ │ "project-b" │
└───────────────┘ └───────────────────┘
Components
1. Central MCP Server (vscode_mcp_server.py)
Python server that:
- Exposes MCP endpoint at
http://localhost:9876/mcp - All tools take
workspaceas first argument - Routes tool calls to registered VS Code extensions
- Manages extension lifecycle (heartbeat, cleanup)
Run:
cd vscode
python vscode_mcp_server.py # Default port 9876
python vscode_mcp_server.py --port 9999 # Custom port
2. VS Code Extension (extension/)
TypeScript extension that:
- Starts local HTTP server to receive tool calls
- Registers with central server on activation
- Uses VS Code API to manage terminals
- Tracks terminal output for reading
Install:
cd vscode/extension
npm install
npm run compile
# Then use "Extensions: Install from VSIX..." in VS Code
Usage
Agent Configuration
{
"mcpServers": {
"vscode": {
"url": "http://localhost:9876/mcp"
}
}
}
Available Tools
All tools take workspace as the first argument:
terminal_create
Create a new terminal in a workspace.
terminal_create(
workspace="my-project",
name="Agent Terminal",
cwd="/path/to/project" # optional
)
# Returns: terminal_id (string)
terminal_exec
Execute a command in a terminal.
terminal_exec(
workspace="my-project",
terminal_id="term_abc123",
command="ls -la"
)
# Returns: "Executed"
terminal_read
Read terminal output since last read.
terminal_read(
workspace="my-project",
terminal_id="term_abc123",
since_index=0 # 0 = all, use returned next_index for incremental
)
# Returns: JSON string with output and next_index
terminal_list
List all active terminals in a workspace.
terminal_list(workspace="my-project")
# Returns: JSON array of terminal info
terminal_kill
Kill a terminal.
terminal_kill(
workspace="my-project",
terminal_id="term_abc123"
)
# Returns: "Killed"
Example Workflow
# Create terminal
term_id = terminal_create(workspace="my-project", name="Build")
# Execute command
terminal_exec(workspace="my-project", terminal_id=term_id, command="npm install")
# Read output
result = terminal_read(workspace="my-project", terminal_id=term_id, since_index=0)
data = json.loads(result)
print(data['output']) # Terminal output
next_index = data['next_index']
# Read more output later
result = terminal_read(workspace="my-project", terminal_id=term_id, since_index=next_index)
Policy Proxy Integration
Add to your policy configuration:
# dev/config/mcp-gateways/policy/vscode.yaml
backend:
name: vscode
url: http://localhost:9876
path: /mcp/vscode
transport: http
rules:
# Deny dangerous commands
- match:
tool: "terminal_exec"
command: "^(rm|sudo|chmod|chown|dd|mkfs)\\b"
action: deny
reason: "Dangerous command blocked"
# Require confirmation for git force operations
- match:
tool: "terminal_exec"
command: "git (push --force|reset --hard|rebase)"
action: confirm
reason: "Destructive git operation requires approval"
# Allow read-only commands
- match:
tool: "terminal_exec"
command: "^(cat|less|head|tail|grep|ls|echo|pwd|whoami)\\b"
action: allow
# Default deny
- match:
tool: ".*"
action: deny
reason: "No matching policy rule"
Agent connects through policy proxy:
{
"mcpServers": {
"vscode": {
"url": "http://policy-proxy:8000/mcp/vscode"
}
}
}
Extension Settings
vscode-mcp.serverUrl: URL of central MCP server (default:http://localhost:9876)vscode-mcp.heartbeatInterval: Heartbeat interval in seconds (default: 30)vscode-mcp.localPort: Local port for extension server (default: 0 = random)
API Reference
Central Server Endpoints
POST /register
Register a VS Code extension.
Request:
{
"workspace": "my-project",
"extension_url": "http://localhost:9877"
}
Response:
{
"status": "registered",
"workspace": "my-project",
"endpoint": "/mcp"
}
POST /heartbeat
Send heartbeat to keep registration alive.
Request:
{
"workspace": "my-project"
}
Response:
{
"status": "ok"
}
GET /workspaces
List all registered workspaces.
Response:
{
"workspaces": [
{
"name": "my-project",
"url": "http://localhost:9877",
"last_heartbeat_seconds_ago": 5.2,
"registered_at": "2024-01-01T12:00:00"
}
]
}
GET /health
Health check.
Response:
{
"status": "ok",
"extensions_count": 2,
"workspaces": ["my-project", "other-project"]
}
Extension Local Server Endpoints
POST /execute
Execute a tool via VS Code API.
Request:
{
"tool": "terminal_create",
"arguments": {
"name": "My Terminal",
"cwd": "/path/to/project"
}
}
Response:
{
"result": "term_abc123"
}
Installation
From PyPI (recommended)
pip install vscode-mcp
From GitHub
pip install git+https://github.com/oscaryard/vscode-mcp
From source
git clone https://github.com/oscaryard/vscode-mcp
cd vscode-mcp
pip install -e .
Run the server
# Using the installed command
vscode-mcp-server --port 9876
# Or using Python module
python -m vscode_mcp --port 9876
Development
Setup Central Server
cd vscode
pip install -r ../../mcp/requirements.txt
python vscode_mcp_server.py
Setup Extension
cd vscode/extension
npm install
npm run compile
Press F5 in VS Code to launch Extension Development Host.
Troubleshooting
Extension not registering:
- Check central server is running:
curl http://localhost:9876/health - Check extension logs: View → Output → VS Code MCP Extension
- Check central server logs for registration attempts
Tool calls failing:
- Verify workspace name matches:
curl http://localhost:9876/workspaces - Check extension is running: Look for status bar icon
- Check local server is running: Extension logs should show port
Terminal output not captured:
- Output is only captured after extension activates
- Buffer limited to last 1000 lines
- Terminal must be created via MCP tools (not manually)
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。