Drop WPLoadTester into your agentic CLI.
Already running Claude Code, Codex, OpenCode, Cursor, or Windsurf against your codebase? Point it at WPLoadTester too. The MCP server in WPLoadTester 7 exposes 75+ tools (test-case configuration, ASM correlation, load execution, error triage, post-test analysis) to any MCP-compatible client. Localhost, auto-starts with the app, no extra license.
Watch your AI debug a failed load test.
Triaging a failed load test today is the same loop every senior performance engineer has run a hundred times: open the Errors view, copy the URL, recognize the pattern (or not), apply the fix, rerun. With the MCP server, the agent you already use for code can run that loop too. The same Claude Code session you use to refactor a module can now read your replay errors, recognize a Reblaze anti-bot challenge, call the ASM correlation tool, and propose the fix, in the same terminal with the same approval flow you already trust. Mutation tools require confirmation through your client's normal approval gate; reads are unprompted.
A battle-tested expert system underneath the conversation.
Asking an LLM to correlate hundreds, or thousands, of dynamic fields across the pages of a real test case is a recipe for hallucination. Wrong extractors. Missed session tokens. Replays that look correct in the AI's summary but fail the moment they hit the server. WPLoadTester doesn't do that.
Underneath the AI is a deterministic expert system encoded from thousands of real test cases. Every auth pattern (PKCE, OAuth, SAML, classic form login), every common framework's dynamic-field convention, every variation of session state we've encountered in 20+ years of consulting work. The rules are unit-tested. They run identically every time. They don't invent extractors that don't exist.
The AI handles the open-ended part: reading your specific test case, deciding which rules apply, explaining its reasoning, and handling the edge cases the rules don't cover. The expert system handles the precision work, where being right matters more than being interesting. That's why this scales to test cases that a generic LLM tool can't touch without falling over.
Three stages, one assistant.
From recording through analysis, the AI uses the same tools you'd reach for. Faster, and without you switching panels.
Configure
Hand the AI a recorded session. It walks the test structure, runs ASM correlation, creates the extractors it decides it needs (boundary, regex, JSON path), and iterates until replay passes, surfacing each attempt instead of guessing silently.
Try asking
- "Look at my test case and tell me if ASM correlation looks complete."
- "My replay is failing with 403 errors. What's wrong?"
Monitor
While a load test is running, the AI reads live response time, throughput, and error rates straight from WPLoadTester, then tells you what it's seeing in plain English instead of making you watch a dashboard.
Try asking
- "Start a load test with 100 users ramping over 5 minutes."
- "How does response time look at the current user level?"
Analyze
When the test finishes, the AI ranks pages by total time impact, finds inflection points in the user-level curve, correlates page slowdowns with server CPU/memory, and compares error patterns between runs.
Try asking
- "Which pages consume the most total time?"
- "Compare error rates between 50 and 100 users."
Auto-starts on localhost. No setup.
WPLoadTester runs the MCP server automatically whenever the application is open. Default port 9100, localhost-only, and it auto-increments through 9109 if the default port is in use. There is no "enable MCP" toggle and no key to manage. If a process on your machine can reach localhost, it can already authenticate as you, so the server inherits that trust boundary.
Connect Claude Code with one line:
claude mcp add loadtester --transport http http://localhost:9100/mcp
Or, for Claude Desktop, add this to claude_desktop_config.json:
// claude_desktop_config.json { "mcpServers": { "loadtester": { "url": "http://localhost:9100/mcp" } } }
Works with Claude Desktop, Claude Code, Codex, OpenCode, Cursor, Windsurf, and any other MCP-compatible client. The docs cover the full client matrix and per-client configuration notes. See the client setup guide
75+ tools across the lifecycle.
The server exposes 75+ tools spanning test-case structure, ASM correlation, extractor creation, dataset configuration, load-profile setup, live monitoring, error filtering, page-level performance ranking, server-resource correlation, and cross-test comparison. Every tool is read-safe by default; anything that changes state runs through your client's normal approval prompt.
Try it against your own test cases.
The MCP server is included with every WPLoadTester 7 install. Request the beta, drop the connection line into your AI client, and run your existing agentic workflow against your load tests.