MCP Server¶

Load Tester runs an MCP server automatically whenever the application is open. You do not need to enable it, configure it, or think about it.

MCP (Model Context Protocol) is a standard protocol that lets AI assistants interact with external tools. The server exposes 75+ tools covering test case configuration, load test monitoring, and performance analysis. Any AI tool that supports MCP can connect: Claude Desktop, Claude Code, Cursor, Windsurf, or anything else that speaks the protocol.

This means you can use your preferred AI tool to work with Load Tester, not just the embedded panel. Ask questions in natural language and the AI figures out which Load Tester tools to call.

How It Works¶

Load Tester starts an embedded Jetty HTTP server on localhost when the application launches. Nothing to configure.

Detail	Value
URL	`http://localhost:9100/mcp`
Default port	9100 (auto-increments to 9101-9109 if occupied)
Protocol	MCP Streamable HTTP (JSON-RPC 2.0)
Binding	Localhost only (127.0.0.1)

The server binds to 127.0.0.1, which means it only accepts connections from your own machine. Nothing on the network can reach it. Authentication is unnecessary because if a process can connect, it is already running on your computer with your credentials.

If port 9100 is already in use (another instance of Load Tester, or a different application), the server tries 9101, then 9102, up through 9109. Check the console output if you need to confirm which port it landed on.

Connecting External AI Tools¶

Claude Desktop¶

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "loadtester": {
      "url": "http://localhost:9100/mcp"
    }
  }
}

Restart Claude Desktop after saving the file. Load Tester tools appear automatically once both applications are running.

Config File Location

On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

On Windows: %APPDATA%\Claude\claude_desktop_config.json

Claude Code¶

Run this from a terminal:

claude mcp add loadtester --transport http http://localhost:9100/mcp

That's it. The tools are available immediately in your next Claude Code session.

Other MCP Clients¶

Any MCP-compatible client can connect to http://localhost:9100/mcp using the Streamable HTTP transport. Cursor, Windsurf, and similar tools all support MCP configuration. Point them at the URL and the tools register themselves.

Available Tool Categories¶

The MCP server exposes tools organized around the three stages of load testing.

Test Case Configuration (Stage 1)¶

Analyze test case structure and transactions
Run ASM correlation wizards
Create and configure extractors (boundary, regex, JSON path)
Manage detection rules
Configure datasets and field assignments
Debug replay failures and error patterns

Load Test Management¶

Discover and configure load profiles
Set user levels and ramp-up patterns
Start and stop load tests
Monitor test execution status

Live Monitoring (Stage 2)¶

Get real-time metrics: response time, throughput, error rates
Track error timeline and patterns
Monitor page-level performance trends
View metrics broken down by user level
Detect performance degradation as it happens

Performance Analysis (Stage 3)¶

Filter errors by HTTP status code
Rank pages by cumulative time impact
Correlate page performance with server metrics
Detect inflection points and degradation between user levels
Analyze scaling patterns
Compare error patterns across time windows

Server Monitoring¶

Access server CPU, memory, disk, and network metrics
Correlate server resource usage with application performance

What You Can Do With MCP¶

The MCP connection is open-ended. Describe what you want in natural language and the AI handles the tool calls. You do not need to know which tools exist or how to call them.

From Claude Code or Claude Desktop, you can ask things like:

"Look at my test case and tell me if the ASM correlation looks complete"
"My replay is failing with 403 errors. What's wrong?"
"Start a load test with 100 users ramping up over 5 minutes"
"The load test is running. How does response time look at the current user level?"
"Analyze the completed load test and identify the bottleneck"
"Compare error rates between 50 and 100 users"
"Which pages consume the most total time?"

The AI reads your test case, load test results, and live metrics directly from Load Tester. It calls whichever tools it needs, chains them together, and gives you a coherent answer. You just ask the question.

This also means you can build workflows that combine Load Tester data with other sources. Ask Claude Code to analyze your load test results alongside your application logs, or have it cross-reference error patterns with your Git history. The MCP tools give the AI access to Load Tester data; the AI can combine that with anything else it can reach.

Security¶

Localhost only. The MCP server binds to 127.0.0.1 and rejects connections from any other address. Nobody on your network can reach it.

No authentication required. Only local processes can connect. If a process can reach the server, it is already running on your machine with your credentials. Adding a password would protect nothing.

Read-heavy, confirmation for writes. Most tools read test data: test case structure, metrics, errors, configuration. Tools that modify state (starting a load test, changing configuration) require confirmation through the AI client's normal approval flow.

Scoped to the current session. Tools operate on whatever test case or load test is currently open in Load Tester. They cannot access files on disk, other applications, or data from previous sessions.

Troubleshooting¶

"Cannot connect to MCP server"¶

Make sure Load Tester is running. The MCP server only exists while the application is open.

Port conflict¶

If another application is using port 9100, Load Tester automatically tries the next port (9101, 9102, etc.). Check the console output for: MCP Server: started on port N. Update your MCP client configuration to match the actual port.

Tools not appearing in the AI client¶

Verify the MCP client config points to the correct URL. Make sure both Load Tester and the AI client are running. Some clients (Claude Desktop) require a restart after changing the MCP configuration.

"No active test case"¶

Open a test case in Load Tester before asking configuration questions. The AI tools need a test case loaded to analyze.

"No active load test"¶

Start a load test or open a saved load test result before asking monitoring or analysis questions. Without an active or completed load test, the monitoring and analysis tools have nothing to report on.