Creativity · MCP — pattern
Testing MCP Servers with the Inspector Tool
`@modelcontextprotocol/inspector` is the official test harness for MCP servers. You point it at your server command (or URL) and it launches a browser UI where you can list tools, call them with JSON arguments, stream resources, and watch the JSON-RPC traffic live. Treat it as your curl+Postman equivalent for MCP. It's the fastest way to verify a new server works before you wire it into Claude Desktop.
MCP facts
- Kind
- pattern
- Ecosystem
- anthropic-mcp
- Language
- TypeScript / Node.js
- Transports
- stdio, http, sse
Capabilities
- List tools, resources, prompts exposed by a server
- Invoke tools with structured JSON arguments
- Subscribe to resources and see update notifications
- Inspect raw JSON-RPC request/response traffic
Install
npx -y @modelcontextprotocol/inspectorConfiguration
# Inspect a local stdio server
npx -y @modelcontextprotocol/inspector npx -y @modelcontextprotocol/server-filesystem /tmp
# Inspect a remote HTTP server
npx -y @modelcontextprotocol/inspector --transport http --url https://mcp.example.comFrequently asked questions
When should I use the Inspector?
Any time you build or modify an MCP server. It's faster than restarting Claude Desktop, and you can see the exact JSON-RPC payloads — great for debugging schema mismatches.
Does it support auth?
Yes. For HTTP/SSE servers you can provide a bearer token or trigger the OAuth flow directly in the UI. For stdio servers, pass env vars on the command line the usual way.
Can I script it for CI?
The UI is interactive, but the underlying MCP SDK libraries are easy to script for automated tests. A common pattern is: Inspector for dev, SDK-based integration tests in CI.
Sources
- MCP Inspector on GitHub — accessed 2026-04-20
- MCP docs — debugging — accessed 2026-04-20