@browserstack/mcp-server 1.5.0-beta.20 → 1.5.0-beta.22

package/capability/loadtesting.capability-index.json

@@ -564,8 +564,9 @@
         ],
         "intent": "Get the AI insight (summary, root cause, recommendations) for a completed run — use this for 'why did this run regress or fail?'",
         "guidance": [
+          "This is the canonical answer to a 'summarize this run' / 'give me a summary' request: return ONLY these AI insight summary points and stop. Do NOT also call getLoadTestRunReport or append KPIs / web vitals / transactions unless the user explicitly asks for them.",
           "Only meaningful after the run is terminal; a running job returns an in_progress state, not a report.",
-          "May return 503 if the insight backend is unavailable — retry later."
+          "May return 503 if the insight backend is unavailable — retry later. If no insight exists for the run, fall back to a brief status + headline KPIs from getLoadTestRunReport, not the full report."
         ],
         "returns": [
           "state",
@@ -681,7 +682,8 @@
           "groupBy and errorCategory narrow the payload; byteCap hard-caps the response (max 256 KB).",
           "metrics accepts dotted names or @ aliases; see getLoadTestMetricsManifest for what applies to this test type.",
           "Answer a scoped question with the narrowest slice instead of the whole report: one metric → metrics=; error breakdown → errorCategory= (or slaOnly=true); a time window → sinceIso/untilIso; slowest transaction → detail=per-txn with groupBy=transaction and topN.",
-          "One call with the right params returns everything for that question — do not re-fetch the same run with the same params. Fetch the full report (view=full / detail=full) only when the user explicitly asks for the raw or complete report."
+          "One call with the right params returns everything for that question — do not re-fetch the same run with the same params. Fetch the full report (view=full / detail=full) only when the user explicitly asks for the raw or complete report.",
+          "A plain 'summarize this run' / 'give me a summary' request is answered by getLoadTestInsightSummary, NOT this endpoint. Only call this when the user asks for specific KPIs, SLA verdicts, web vitals, transactions, or the raw/complete report."
         ],
         "returns": [
           "runId",
@@ -938,7 +940,8 @@
             "name": "name",
             "type": "string",
             "required": true,
-            "description": "Test name."
+            "maxLength": 255,
+            "description": "Test name. Max 255 characters — truncate before sending; a longer name is rejected with a 400."
           },
           {
             "name": "testType",
@@ -1026,7 +1029,10 @@
           "Only THREE things must be known before creating, and each is taken from the user's request when they've stated it — ask ONLY for the ones still missing, and ask about nothing else: (1) testType — plu (protocol / API load), blu (real-browser load) or hybrid (both); (2) framework — k6 / jmeter / gatling / locust for plu, playwright / selenium / webdriverio / nightwatch for blu; (3) the script — a built-in sample (scriptRef:{source:'sample'}; for Selenium also pass scriptRef.subType) OR the user's own script via the pendingScriptUpload flow. Do not guess a framework/testType, but do not re-ask for one the user already gave.",
           "There is no clone capability. To duplicate a test, getLoadTest the source and copy its full config into this create — vuRamp/vus, durationSec, loadGeneratorLocations and slaThresholds included; nothing is inherited from the source, so anything you omit is dropped.",
           "DEFAULTS apply ONLY after the three mandatory fields (testType, framework, script) are known — they never substitute for a missing one of those. Everything BEYOND those three — do not ask about it; apply the default unless the user explicitly specifies otherwise: loadProfile='constant', config.vus=1, config.durationSec=30, region us-east-1 (omit loadGeneratorLocations so the backend defaults it), and NO advanced features (do not enable log capture, response capture, artifacts, thresholds, etc.). Send config.vus=1 + config.durationSec=30 with no vuRamp for the default constant profile.",
-          "Only when the user explicitly asks for a different load shape, set config.loadProfile: 'ramping' (config.vuRamp: array of {vus, durationSec} stages; ramp up/down is inferred from adjacent VU deltas), 'iterations' (PLU only — config.iterations = fixed run count, plus config.vus), or 'throughput' (PLU and only JMeter/Gatling/locust/k6 — config.targetRps + config.maxVus + config.durationSec). A stage's own 'type' is NOT honored — every stage persists as a hold."
+          "Only when the user explicitly asks for a different load shape, set config.loadProfile: 'ramping' (config.vuRamp: array of {vus, durationSec} stages; ramp up/down is inferred from adjacent VU deltas), 'iterations' (PLU only — config.iterations = fixed run count, plus config.vus), or 'throughput' (PLU and only JMeter/Gatling/locust/k6 — config.targetRps + config.maxVus + config.durationSec). A stage's own 'type' is NOT honored — every stage persists as a hold.",
+          "testType='hybrid' REQUIRES children[] with the PLU leg (top-level is the BLU leg). NEVER call create with testType='hybrid' and no children[] — it is rejected 'hybrid testType requires children[]'. If you don't yet have both legs' framework+script, ASK for them first; do not submit a hybrid without children to 'see what happens'.",
+          "On a 400, FIX the payload per the error message and only then resend — NEVER resubmit an unchanged payload (add children[] for hybrid, shorten name to <=255, supply testType+framework). Do not call getLoadTest after a FAILED create; there is nothing to fetch. Repeatedly retrying the same rejected body is the main source of wasted create calls.",
+          "Always send idempotencyKey, and reuse the SAME key across retries of one logical create, so a late success can't create a duplicate test."
         ],
         "returns": [
           "testId",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@browserstack/mcp-server",
-  "version": "1.5.0-beta.20",
+  "version": "1.5.0-beta.22",
   "description": "BrowserStack's Official MCP Server",
   "mcpName": "io.github.browserstack/mcp-server",
   "main": "dist/index.js",