v0.4.5

Warns about Java/Groovy mismatches, fixes cross-platform detection, and lets thread groups stagger their start. Supersedes 0.4.4, which was never published to npm.

• Warn clients about the Java/Groovy mismatch before JSR223 is used: the Groovy bundled with JMeter 5.6.x fails on very new Java, and each failing script silently kills that virtual user's iteration. add_jsr223_* results, execute_test_plan, and find_breaking_point now return a warning when the detected Java is newer than the bundled Groovy supports; add_jsr223_* descriptions steer toward ${__UUID}/__RandomString/__Random and state the Java requirement; Groovy is detected under lib/ and Homebrew's libexec/lib.
• Make Java/Groovy detection platform-aware for Windows and macOS: build paths with win32/posix rules per target platform, strip quotes/whitespace from JAVA_HOME/JMETER_HOME, hide the console window when probing java -version on Windows, and unit test both platforms from any host.
• Add delaySeconds start delay to thread groups: ThreadGroup.delay was hardcoded to empty, so clients couldn't stage thread groups. Now serialized and parsed, exposed on the add_*thread_group tools, and cleared during find_breaking_point rounds.

v0.4.3

Adds live progress to get_execution_status.

• While a run is going, get_execution_status now returns samples, error rate, avg/p95 latency, and overall and per-label throughput, computed from the results file so far. Polling clients no longer need to tail stdout.log, whose summary line only refreshes every ~30s.

v0.4.2

Fixes a request mix skew bug and improves breaking point reporting.

• Fix: Nested Loop Controllers were serialized with continue_forever=false, so they marked themselves done after one pass and were skipped on later iterations of a looping/scheduler thread group, skewing request mix (e.g. 604 logins vs 40 signups instead of the intended ratio). Now emits continue_forever=true, matching the JMeter GUI.
• Fix: avoid a stack overflow in aggregate metrics on large sample sets.
• find_breaking_point: per-label metrics per round, breakingPointRange with an explicit exact flag, progress and file paths in status.
• get_execution_status: filter JVM startup noise, add elapsedSeconds.
• New add_once_only_controller tool.
• Clarify add_http_sampler description; document workspace and mix behavior.

v0.4.0

Adds capacity search for finding a test plan's breaking point.

• find_breaking_point, get_breaking_point_status, stop_breaking_point_search: three async tools that find the concurrency level where a plan stops meeting its SLA.

v0.3.1

Fix: .jmx import was silently resetting a Thread Group's loop count to 1. Docs: clarified that CSV path validation only runs at add_csv_data_set time.

v0.3.0

New (minor): test plan editing tools — remove_element, update_element, rename_element, move_element, reorder_children, set_element_enabled — and .jmx import via import_test_plan. Also fixes a Windows-only spawn EINVAL crash in execute_test_plan.

v0.2.2

Fix: JTL parser was corrupting rows containing multi-line quoted fields.

v0.1.2

Fix: invalid guiclass on the Aggregate Report listener that could produce a .jmx JMeter wouldn't load correctly.

v0.1.1

Docs: expanded the README with the rationale behind the project, and translated the docs to English.

v0.1.0

Initial release: stdio MCP server for building Apache JMeter test plans, running load tests, and reading back execution reports.