MaxoPerf agents — build and run tests from any AI coding agent

MaxoPerf is built to be driven by an AI agent, not just a human at a keyboard. Any agent that speaks MCP can call MaxoPerf’s tools directly — but the MaxoPerf skill goes further: it’s a small, shippable instruction bundle that teaches a skill-aware agent (Claude Code, Claude Desktop, and compatible tools) the actual MaxoPerf workflow — how to pick a test type, where secrets go, what a healthy run looks like — the knowledge that would otherwise take a long, error-prone prompt to explain every time.

What the skill teaches an agent

The skill turns “I have an app” into a real, uploaded, running MaxoPerf test through a fixed sequence, grounded entirely in shipped platform facts (never invented):

1. Intake, one question at a time. What to test, what the target is (a live URL, or something local/undeployed — in which case the agent offers a reverse tunnel), roughly how much load, and what counts as pass/fail.
2. An optional hotspot scan. If asked, the agent can grep the user’s own repo for likely load-testing hotspots (heavy routes, N+1 queries, fan-out calls) and offer a short, encouraging shortlist — read-only, nothing ever leaves the agent’s own context.
3. Choosing an executor, grounded in the platform’s real supported-executor list — k6 or jmeter for API/protocol targets, playwright or selenium for browser journeys, or reusing whatever asset the user already has (a JMX file, a Locust script, a .side recording). See Tests & builders for the full picture of what MaxoPerf can run.
4. Building the test artifact in the shape the platform actually accepts — for browser tests specifically, every supported script shape (Playwright spec/library/IIFE, Selenium test-function/bare script, or a scriptless interactive journey) so the agent never rewrites a user’s existing script into something it doesn’t have to be.
5. Uploading and running it via the MCP tools — create_test, upload_test_file, start_run — then polling to a terminal state.
6. Reading results back, including a console deep link, and — on a failed or suspicious run — root-causing it with the diagnostics tools (get_run_errors, get_run_logs, get_runner_health, detect_run_anomalies).
7. API-key onboarding the first time an agent hits an auth error, so a fresh session can get connected without leaving the conversation.

Multi-environment awareness (staging vs. production, or per-region targets) and platform features beyond the core loop — tunnels, browser fleets, virtual services, and on-demand browser inspection for single-page apps — are part of the same skill, reached for exactly when the situation calls for them, not by default.

Zero invented facts, by design

Every executor name and tool argument the skill uses is pinned to MaxoPerf’s actual shipped source — the skill’s reference pages are the same ground truth Loadrigo and this documentation site are grounded in. If a capability isn’t real, the skill doesn’t claim it.

Install the skill

1. Get a MaxoPerf API key first — see Get an API key. Keep it handy; the installer asks for it.

2. Run the installer from a terminal where your agent (Claude Code, etc.) is available:

   npx @maxoperf/agent-skill install

3. Paste your API key when prompted. It’s stored the same way a manual MCP config would store it — never printed back out or logged.

4. Restart or reload your agent so it picks up the new skill and MCP connector.

The installer is idempotent — re-run it any time to update the skill in place. Full walkthrough (including uninstall and first-run prompts to try): Install the MaxoPerf skill.

No skill support? Connect over MCP directly

If your agent doesn’t support skills, it can still do everything above by calling MaxoPerf’s tools directly once connected — see Connect your AI agent to MCP for per-client setup and Tools for the full catalog. Chat-only clients that can’t read a filesystem (Claude Desktop, ChatGPT) get the same intake → build → run flow through the server’s guided prompts instead.

Safety

An API key — whether wired into a manual MCP config or stored by the skill installer — grants full-account access. Review your agent’s tool-call log before approving create/run/delete actions, especially the first few times, and treat the connection like any other credential. See Get an API key for scoping and rotation guidance, and read-only mode if you want an agent to investigate without being able to change anything.

Where to go next

MCP overview The endpoint, authentication, and read-only mode underneath every agent connection.

MCP tools reference Every tool, prompt, and resource the skill and any other agent can call.

Loadrigo MaxoPerf's own in-console assistant, built on the same tools and skill.

Tests & builders What kinds of tests an agent (or you) can build — performance, functional API, browser, scriptless.