RayBridge
Bridges locally installed Raycast extensions to any MCP-compatible client via stdio or HTTP. It enables headless execution of extension logic by shimming UI components and leveraging existing OAuth tokens.
README
RayBridge
MCP server that bridges Raycast extensions to any MCP-compatible client.
Discovers locally installed Raycast extensions, loads their tool definitions, and serves them over the Model Context Protocol via stdio or HTTP.

How it works
- Scans
~/.config/raycast/extensions/for installed extensions withtoolsdefinitions - Loads OAuth tokens from Raycast's encrypted SQLite database
- Registers tools as MCP tools accessible to any MCP client
Extensions that use Raycast UI APIs (List, Detail, Form, etc.) are supported — the UI components are shimmed to no-ops so the underlying tool logic can execute headlessly. Extensions whose tools perform background work (API calls, data lookups, transformations) work best.
Setup
Prerequisites
- Bun
- Raycast installed with extensions
sqlcipherCLI (for OAuth token access):brew install sqlcipher
Install
bun install
Configure MCP client
Claude Code (~/.claude/settings.json):
{
"mcpServers": {
"raybridge": {
"command": "bun",
"args": ["run", "src/index.ts"],
"cwd": "/path/to/raybridge"
}
}
}
Cursor (~/.cursor/mcp.json):
{
"mcpServers": {
"raybridge": {
"command": "bun",
"args": ["run", "src/index.ts"],
"cwd": "/path/to/raybridge"
}
}
}
HTTP Transport
The server can also run as an HTTP server for remote MCP clients.
Start the server:
# Default: http://0.0.0.0:3000
bun run start:http
# Custom host/port
MCP_PORT=8080 MCP_HOST=0.0.0.0 bun run start:http
# With API key authentication
MCP_API_KEY=your-secret-key bun run start:http
# CLI flags also work
bun run src/index.ts --http --port 8080 --host 0.0.0.0
Endpoints:
| Endpoint | Method | Description |
|---|---|---|
/health |
GET | Health check (no auth required) |
/mcp |
POST | MCP requests (requires auth if MCP_API_KEY set) |
/mcp |
DELETE | Terminate session |
Authentication:
When MCP_API_KEY is set, requests to /mcp must include a Bearer token (per MCP spec):
Authorization: Bearer your-secret-key
Example session:
# 1. Initialize session (capture session ID from response header)
curl -X POST http://127.0.0.1:3000/mcp \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-H "Authorization: Bearer your-secret-key" \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{
"protocolVersion":"2024-11-05",
"capabilities":{},
"clientInfo":{"name":"my-client","version":"1.0"}
}}'
# Response includes: mcp-session-id header
# 2. List available tools
curl -X POST http://127.0.0.1:3000/mcp \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-H "Authorization: Bearer your-secret-key" \
-H "mcp-session-id: <session-id-from-step-1>" \
-d '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}'
# 3. Call a tool
curl -X POST http://127.0.0.1:3000/mcp \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-H "Authorization: Bearer your-secret-key" \
-H "mcp-session-id: <session-id-from-step-1>" \
-d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{
"name":"web",
"arguments":{"tool_name":"read_page","input":{"url":"https://example.com"}}
}}'
Sessions auto-expire after 30 minutes of inactivity.
CLI
RayBridge includes a CLI for managing which extensions and tools are exposed:
bun link # Register the raybridge command (one-time setup)
raybridge # Launch interactive TUI
raybridge config # Launch interactive TUI
raybridge list # List all extensions and their status
raybridge help # Show help
The TUI allows you to:
- Toggle extensions on/off
- Expand extensions to toggle individual tools
- Switch between blocklist mode (all enabled by default) and allowlist mode
- Save configuration to
~/.config/raybridge/tools.json
Configuration
Tools configuration
Control which extensions and tools are exposed via ~/.config/raybridge/tools.json:
{
"mode": "blocklist",
"extensions": {
"extension-name": {
"enabled": false
},
"another-extension": {
"enabled": true,
"tools": ["specific-tool-1", "specific-tool-2"]
}
}
}
- blocklist mode (default): All extensions enabled unless explicitly disabled
- allowlist mode: All extensions disabled unless explicitly enabled
Extension preferences
Extensions that require configuration (API keys, personal access tokens, etc.) read from:
~/.config/raybridge/preferences.json
{
"extension-name": {
"personalAccessToken": "your-token",
"apiKey": "your-key"
}
}
The extension name matches the name field in the extension's package.json.
Architecture
src/
├── index.ts # MCP server, tool registration, request dispatch
├── http-server.ts # HTTP transport with session management
├── cli.ts # CLI entry point (config, list, help commands)
├── tui.tsx # Interactive TUI for extension configuration
├── config.ts # Tools configuration (blocklist/allowlist)
├── discovery.ts # Scans ~/.config/raycast/extensions/ for tool definitions
├── loader.ts # Executes local tools with Raycast API shims
├── shims.ts # Fake @raycast/api, react, react/jsx-runtime modules
├── auth.ts # Keychain access, SQLcipher DB decryption, OAuth tokens
└── watcher.ts # Watches extension directories for changes, triggers reloads
Tool discovery
Local extensions are discovered from ~/.config/raycast/extensions/. Each extension's package.json must have a tools array defining available tools with names, descriptions, and input schemas. Compiled tool code lives at tools/{toolName}.js within each extension directory.
When duplicates exist (same extension name in multiple directories), the most recently modified version wins.
Tool execution
Tools are loaded by installing Raycast API shims into Node's module system, then requiring the tool's compiled JS file and calling its default export with the provided input.
Raycast API shims
The following @raycast/api features are shimmed:
| Feature | Behavior |
|---|---|
OAuth.PKCEClient |
Returns tokens from Raycast's encrypted DB |
getPreferenceValues() |
Returns values from preferences.json |
environment |
Provides extension name, paths, version info |
Cache |
In-memory key-value store |
showToast, showHUD |
No-op (logs to stderr in some cases) |
open, closeMainWindow, popToRoot |
No-op |
confirmAlert |
Returns undefined |
UI components (List, Detail, Form, etc.) |
Return null |
LocalStorage |
No-op |
Clipboard |
No-op |
React and JSX runtime are also shimmed with minimal mocks (createElement → null, hooks are no-ops).
Authentication
OAuth tokens are read from Raycast's encrypted SQLite database at:
~/Library/Application Support/com.raycast.macos/raycast-enc.sqlite
The database key is retrieved from macOS Keychain and derived with a salt via SHA256. Tokens are extracted per-extension and provided to tools through the OAuth.PKCEClient shim.
MCP tool schema
Extensions are grouped — each extension becomes one MCP tool. The input schema follows this pattern:
{
"tool_name": "which-tool-to-run",
"input": { "param": "value" }
}
Tool descriptions include per-tool documentation, parameter details, and any extension-wide AI instructions from the extension's ai.instructions field.
Limitations
- No interactive UI — extensions that depend on rendering Lists, Forms, or other visual components to the user won't behave meaningfully
- No persistent LocalStorage — shimmed as no-op; extensions relying on it lose state between calls
- OAuth tokens are not refreshed — expired tokens will cause failures until Raycast refreshes them
- macOS only — depends on macOS Keychain and Raycast's macOS app paths
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。