MCP Inspector is a developer tool for exploring, invoking, and diagnosing Model Context Protocol servers before connecting them to an agent. Think of it as a live test bench: connect to a server, inspect the tools exactly as the agent sees them, run calls, and examine the results. section 3.2.2 section 3.6
Use the inspector to check:
- The live tool list, descriptions, parameters, and return types. section 3.2.2
- Whether arbitrary arguments produce the expected response shape. section 3.2.2
- Raw JSON-RPC traffic and initialization errors that an agent client may hide. section 3.2.2
- Whether a decorator-generated schema matches your intention and makes a tool discoverable. section 3.2.2
Inspecting a research server
Start the inspector with an absolute server-file path, open its local URL, select the Tools page, choose get_research_sources, and press Run Tool. The interface displays the output and keeps an operation history. A local server may use STDIO, while an HTTP or SSE server uses a different startup configuration; the package runner does not determine the transport. section 3.2.2 section 3.5
Listing 3.3 Inspecting an MCP server
mcp dev {absolute path to file}01_claude_mcp_server.py #1
OUTPUT
Starting MCP inspector...
Proxy server listening on port 6277
MCP Inspector is up and running at http://127.0.0.1:6274 
#2
Open the runnable command and local interface flow for starting MCP Inspector against a server file. section 3.2.2
