MCP tools
The VegaLoad MCP server has nine tools. Each one is a thin wrapper around a command line command. vegaload doctor checks that your agent host sees all nine.
| Tool | What it does | Same as |
|---|---|---|
list_protocols | Lists the 14 protocols, with options and examples. The agent calls it before it picks a protocol. | vegaload protocols |
create_scenario | Writes a new scenario file. JavaScript by default, or Python. | vegaload new |
generate_from_spec | Writes a scenario and a runbook from a JSON OpenAPI file. Each operation is a named step. YAML is not supported. | vegaload new -from-openapi |
validate_scenario | Runs a scenario once with one user. Returns valid, and the stage and error if not. It makes real calls. | vegaload validate |
run_test | Runs a load test, from a scenario file or from a target and a protocol. Returns the result and the path to an HTML report. | vegaload run |
get_results | Reads back a JSON report from an earlier run. | |
suggest_thresholds | Suggests p95 and error rate limits from a baseline report. | vegaload diagnose -no-llm |
diagnose_failure | Explains a run with rule-based findings. Adds a plain-English summary only if you set VEGALOAD_LLM_PROVIDER. | vegaload diagnose |
compare_reports | Compares a new report with a baseline. Returns the changes and whether p95, error rate or a check got worse. | vegaload compare |
About run_test
run_test takes either scenario_path, or target plus protocol.
It also takes the load shape (vus, duration, executor, stages, rate), thresholds, baseline_path, data files and environment values, and where to write the HTML, JSON and JUnit reports. Example thresholds: "p95 < 300ms", "error_rate < 1%", "p95{protocol=\"ftp\"} < 500ms".
It also takes live, no-live and no-open, and stream_file, stream_push, run_id and labels. By default the run is shown live in the user's browser. The first text item of the answer is always the result JSON. The live view address comes as a second text item, live view: URL, and the agent should tell the user. There is no argument for a push token. See Live view and streaming.
A threshold breach is a normal result with thresholds_passed set to false. It is not a tool error. From a terminal, the same run exits 3.
A target that is not localhost is refused unless it is in allow_targets, or yes is true. See Safety.
About validate_scenario
Call it after you write or edit a scenario, and before run_test. A scenario that is not valid is a normal result. Read valid first. A failed check() does not make a scenario invalid.
Where the list comes from
The tool names and descriptions on this page are from VegaLoad 0.9.0. You can see the live list on your own machine:
vegaload doctor
Doctor prints the nine tool names from a real handshake with the server.