Opens the inspector in your browser at http://localhost:8080
3. Auto-mounted by mcp-use dev
mcp-use includes the prebuilt Inspector package. The development CLI mounts it at /mcp/inspector without a separate project dependency:
import { MCPServer } from "mcp-use";
const server = new MCPServer({
name: "my-server",
version: "1.0.0",
});
// Add your tools, resources, prompts...
export default server;
Run mcp-use dev; the inspector is available at
http://localhost:3000/mcp/inspector.
The UI assets, proxy, and OAuth backend all come from the installed package; no CDN is used. Production server.listen() and mcp-use start do not mount the Inspector.
📖 Usage Guide
Embedding the React client
Applications that import components from @mcp-use/inspector/client must also
load the Inspector theme contract from their global Tailwind stylesheet:
The stylesheet supplies the surface, shadow, hover, and active tokens used by
portaled dialogs, dropdowns, selects, and other Inspector UI primitives. The
standalone Inspector bundles these styles automatically.
Dashboard
The main dashboard is your central hub for managing MCP server connections:
Connected Servers Panel (left): View all your connected servers
Connect Panel (right): Add new server connections with transport, URL, auth, and headers configuration
Quick Actions: Access settings and clear sessions from the top bar
Adding an MCP Server
To connect to an MCP server:
Open Connect Panel: Click the Connect panel on the right side of the dashboard
Configure Transport: The Inspector supports Streamable HTTP only.
Enter Server URL: Input the MCP server endpoint (e.g., https://mcp.linear.app/mcp)
Configure Authentication (if needed): Add OAuth credentials or API headers
Click Connect: Establish the connection
Example URLs:
Linear: https://mcp.linear.app/mcp
Local: http://localhost:3000/mcp
OAuth Authentication
For servers requiring OAuth (like Linear):
After clicking Connect, you'll see the authorization page
Click "Approve" to grant access
The inspector handles the redirect automatically
Server appears in Connected Servers list with a green indicator
Privacy Note: All authentication tokens and credentials are stored securely in your browser's local storage. Nothing is sent to our servers - everything stays on your device.
Connected Servers
Once connected, each server shows:
Connection status indicator (green = connected)
Server name and URL
Available tools count
Action buttons: Inspect, Disconnect, Remove
Click "Inspect" to open the detailed server view.
🔍 Server Detail View
After clicking "Inspect" on a connected server, you'll see four main tabs:
Tools Tab
The Tools tab displays all available tools from the MCP server.
Features:
Browse all available tools with their names and descriptions
Select a tool to view its detailed schema
Click "Execute" to test the tool
Enter JSON parameters in the input panel
View real-time results with syntax highlighting
Support for MCP-UI and OpenAI Apps SDK widgets
Example: When connected to Linear MCP server, you'll see 23+ tools for managing issues, projects, and teams.
Resources Tab
Browse available resources from the MCP server:
View resource descriptions and metadata
Copy resource URIs for use in your applications
Check MIME types and resource properties
Preview resource content
Note: Some servers (like Linear) may not expose resources, in which case you'll see "No resources available".
Prompts Tab
Test and manage pre-configured prompts:
View all available prompts with descriptions
Select a prompt to see its schema
Fill in required arguments in the form
Click "Render" to execute the prompt
Copy the rendered output for use
Note: Not all servers provide prompts. If none are available, you'll see "No prompts available".
Elicitation Tab
The Elicitation tab handles tool requests that require user input during execution.
Features:
View pending elicitation requests in a dedicated queue
Fill and submit form-mode responses directly in the inspector
Accept, decline, or cancel requests
Jump back to tool results after responding
Supported field types:
Text, number/integer, and boolean fields
Single-select enum dropdowns:
type: "string" + enum
type: "string" + oneOf[{ const, title }]
type: "string" + enum + enumNames (legacy)
Multi-select enum groups:
type: "array" + items.enum
type: "array" + items.anyOf[{ const, title }]
This aligns with SEP-1330 enum schema variants used by MCP conformance scenarios.
Chat Tab
The Chat tab provides an interactive interface to test the MCP server with an LLM using BYOK (Bring Your Own Key).
Privacy: Your API key is stored locally in your browser and never sent to our servers. All requests are made directly from your device to your LLM provider.
Setup:
Click "Configure API Key" to open the configuration modal
Select your Provider (OpenAI, Anthropic, Google, or Ollama)
Choose the Model (gpt-4o, claude-3-5-sonnet, etc.)
Enter your API Key (stored locally in browser)
Click "Save Configuration"
For Ollama, the API key is optional and you can point the inspector at a local or remote Ollama host by setting the Base URL (defaults to http://localhost:11434).
Using Chat:
Type natural language queries to test the MCP server
View tool calls in real-time with visual indicators
See detailed JSON output for arguments and results
Watch the assistant's response based on tool results
Track exactly how the MCP server processes requests
⌨️ Keyboard Shortcuts
Shortcut
Action
Cmd/Ctrl + K
Quick search and navigation
Esc
Close modals and overlays
Quick Search (Cmd/Ctrl + K) allows you to:
Quickly connect to a new server
Access documentation and tutorials
Jump to recently used servers
Search and execute tools directly
Navigate to different tabs
💾 Persistency
The inspector automatically saves your configurations:
Server connections persist across page reloads
OAuth sessions persist as AES-256-GCM ciphertext using a non-extractable
origin key in IndexedDB
Secret headers and client secrets are runtime-only and must be supplied
again after reload
Session preferences are maintained automatically
Clear all sessions anytime with the "Clear All" button
Your data never leaves your browser - everything is stored locally for privacy and security.
🐳 Self-Hosting
Deploy the MCP Inspector to your own infrastructure with a single Docker container. Perfect for enterprise environments, air-gapped networks, or when you need full control over your debugging environment.
Docker Image
The official Docker image is available on Docker Hub.
Building the Dockerfile installs the exact Inspector version in package.json.
Select another published version with --build-arg VERSION=x.y.z.
Quick Start:
docker run -d -p 8080:8080 --name mcp-inspector mcpuse/inspector:latest
Then visit http://localhost:8080 to access your self-hosted inspector.
All configuration is optional. The inspector works out of the box with sensible defaults.
Variable
Default
Description
NODE_ENV
production
Node.js environment
PORT
8080
Port to run the inspector on
HOST
0.0.0.0
Host to bind to
📊 Telemetry & Privacy
The MCP Inspector collects anonymized usage data to help us improve the tool.
What Data is Collected?
We collect:
Usage events: When you connect to servers, execute tools, read resources, or call prompts
Anonymous user ID: A randomly generated UUID stored in your browser's localStorage
Technical information: Package version, browser type, and connection types
Performance data: Tool execution duration and success/failure rates
We DO NOT collect:
Personal information (names, emails, etc.)
Actual tool arguments or results
Server credentials or API keys
Resource content or data
How to Opt Out
You can disable telemetry in several ways:
Option 1: Environment Variable
# Set before starting the inspector
export MCP_USE_ANONYMIZED_TELEMETRY=false
Option 2: localStorage (Browser)
// In browser console
localStorage.setItem("MCP_USE_ANONYMIZED_TELEMETRY", "false");
Note: If you previously opted out with mcp_inspector_telemetry_disabled, the inspector automatically migrates that setting to MCP_USE_ANONYMIZED_TELEMETRY=false on startup.
Telemetry Provider
We use PostHog for product analytics and feature usage tracking. PostHog respects your privacy and follows GDPR compliance.
Aún no hay lista de herramientas. Los servidores remotos la reciben tras la primera comprobación de estado.
Preguntas frecuentes
Preguntas
¿Cómo instalo Inspector?
Copia en esta página la configuración stdio para Cursor, Claude Desktop, Claude Code o VS Code. Usa la instalación del publicador, no un proxy de PlugMCP.
¿Inspector necesita una clave de API?
Este registro no indica ninguna variable de entorno obligatoria.
¿Qué transporte usa Inspector?
Inspector se ejecuta localmente por stdio.
¿Cuándo se comprobó Inspector por última vez?
Última comprobación: 2026-10-27. Modern MCP Inspector for remote mcp servers with support for Apps SDK