From a Git repository
Import an existing MCP server from GitHub and deploy it in one click.
Manufact
Product▾
- MCP HostingDeploy MCP apps and servers to production.
- Cross-client testingRun the same checks across ChatGPT, Claude, and more.
- Publishing checksAudit your MCP app against ChatGPT Apps Store and Cloud Connectors requirements.
- Submission PackGenerate the assets you need for marketplace submission.
- Cloud InspectorTrace, replay, and debug MCP traffic in production.
- Public chatEmbeddable chat surfaces for your product.
- AnalyticsUsage, latency, and reliability in one place.
- Products
- Cloud
- mcp-use SDK
- Inspector
- Vibe
Developers▾
- Manufact docsBuild, deploy, and operate on Manufact Cloud.
- SupportCommunity, issue reporting, and premium support.
- TemplatesStart from a deploy-ready MCP framework template.
- MCP Quality ScoreScore your MCP server and browse ranked servers.
- mcp-use docsBuild MCP apps and servers with the mcp-use SDK.
- mcp-use Inspector DocsInspect and debug MCP servers from your browser.
- mcp-use client CLIConnect to and call MCP servers from your terminal.
- MCP server tunnelingExpose a local MCP server through a secure tunnel.
- Pricing
- Blog
Book a demoSign in
How good is your MCP? Get a free quality score.Explore catalog
Trusted by developers debugging MCP servers at top companies
Live demo
Add an MCP server and test it.
Add a remote MCP server and start testing it in the Inspector right here.
Get started
How to use the MCP Inspector
Choose where to run the Inspector, then connect your server and inspect a tool response.
Online
Open the hosted version in your browser. Connect to an HTTP MCP endpoint that your browser or proxy can reach. No installation needed.
inspector.manufact.comnpx
Run locally with a single command. Connect to local servers during development. See the CLI guide for transport and connection options.
npx @mcp-use/inspectorConnect your MCP endpoint
Add a server, choose its transport, and enter the MCP endpoint URL, such as
https://example.com/mcp. Auto mode tries Direct first, then the configured proxy if CORS or network policy blocks the request. Select Direct or Via Proxy when you need to test a specific connection path. Complete the authentication prompt if the server requires it.Run a tool with known inputs
Open Tools, select a tool, and fill in its input form. Start with a read-only operation or a test account. Executing a tool can change data in the connected service.
Inspect the result and request
Check the returned content, structured data, and errors. Open the RPC panel to inspect the JSON-RPC exchange. If the response includes an MCP App, preview its widget and inspect its behavior before testing in the target client.
Run the Inspector with npxConnection and authentication settingsDebug a server that passes Inspector but fails in a client
Testing a server locally? Start here with the open-source Inspector. For production traffic, use Cloud Inspector to monitor MCP calls and inspect request/response traces.
Features
Everything you need to debug MCP servers.
Full visibility into your MCP server during development. Test every primitive before connecting to agents.
Tool testing
List, inspect, and execute MCP tools with custom parameters. See request and response data in real time.
Resource browser
Browse and read resources exposed by your server. View metadata, test subscriptions, and inspect content.
Prompt management
View and test prompt templates with different arguments. Validate prompts before connecting to agents.
RPC logging
See every JSON-RPC message between client and server. Debug exactly what gets sent and what comes back.
Widget preview
Test MCP Apps and MCP-UI widgets directly in the Inspector. See how they render in real chat clients.
Multi-transport
Connect via stdio, SSE, or HTTP. Test your server regardless of how it communicates with clients.
Widget testing
Test MCP Apps with UI widgets.
The Inspector supports MCP Apps and MCP-UI widgets. Test interactive components, verify tool calls, and debug widget state before deploying to production.
What you can test
- Widget rendering: See how your widgets look in real chat clients.
- Tool calls from widgets: Test window.openai.callTool() interactions.
- Display modes: Switch between inline, PiP, and fullscreen.
- Widget state: Inspect and debug widgetState persistence.
- Dev mode: Hot reload for rapid widget development.
- Console proxy: See widget console.log output in the Inspector.
window.openai API emulation
The Inspector emulates the window.openai API used by ChatGPT widgets. Use it to inspect widget behavior, then test in ChatGPT too. An Inspector pass does not prove identical behavior in every client.
// Available in widgets
window.openai.toolInput
window.openai.toolOutput
window.openai.widgetState
window.openai.theme
window.openai.callTool()
window.openai.setWidgetState()
Debugging
See every JSON-RPC message.
The RPC Logger shows all communication between the Inspector and your MCP server. Filter by method, search message content, and debug protocol issues.
Real-time logging
Messages appear as they happen. No refresh needed.
Filter and search
Filter by method type. Search message content.
Message details
Expand any message to see full request and response.
Chat
Chat with your MCP server.
The Inspector includes a chat interface that connects your MCP tools to an LLM. Test how agents will interact with your server in a real conversation.
Bring your own key (BYOK)
The Chat feature uses your own API key. Your key is stored locally in your browser and never sent to our servers. All requests go directly from your device to your LLM provider.
OpenAI, Anthropic, and other providers
Keys stored in browser localStorage
Direct API calls, no proxy
Test tool selection and execution