@fre4x/mcp-inspector

MCP server that lets AI agents connect to, inspect, and test any MCP server at runtime — no browser required.

It wraps the MCP SDK Client directly, so an agent can:

1. Spawn a target MCP server (stdio) or connect to a remote one (SSE / Streamable HTTP)
2. Maintain a live session and call multiple operations
3. Tear down the connection when done

Tools

| Tool | Description | |------|-------------| | inspector_connect | Spawn/connect to an MCP server. Returns session_id. | | inspector_disconnect | Close an active session. | | inspector_list_sessions | List all active sessions with targets. | | inspector_list_tools | List tools exposed by the connected server. | | inspector_call_tool | Call a tool and return its result. | | inspector_list_resources | List resources exposed by the server. | | inspector_read_resource | Read a resource by URI. | | inspector_list_resource_templates | List resource URI templates. | | inspector_list_prompts | List prompts exposed by the server. | | inspector_get_prompt | Retrieve a rendered prompt with arguments. |

Usage

Connecting to a stdio server

inspector_connect(command="node", server_args=["dist/index.js"])
→ { session_id: "session-0" }

inspector_list_tools(session_id="session-0")
→ { tools: [...] }

inspector_call_tool(session_id="session-0", tool_name="hn_get_top_stories", tool_args={ limit: 5 })
→ { content: [...] }

inspector_disconnect(session_id="session-0")

Connecting to a remote server

inspector_connect(server_url="https://example.com/mcp", transport="http")
inspector_connect(server_url="https://example.com/sse", transport="sse",
                  headers={ "Authorization": "Bearer <token>" })

Mock Mode

Run without any real server:

MOCK=true npx @fre4x/mcp-inspector

All tools return realistic fixture data when MOCK=true or INSPECTOR_MOCK=true.

Claude Desktop Configuration

{
  "mcpServers": {
    "mcp-inspector": {
      "command": "npx",
      "args": ["-y", "@fre4x/mcp-inspector"]
    }
  }
}