github MCPJam/inspector v3.6.0
Release v3.6.0

4 hours ago

What's Changed

  • Make electron:dev a usable dev loop by @chelojimenez in #4905
  • Chromium install: keep the reason, retry on its own, skip Electron by @chelojimenez in #4971
  • Docs: document Chromium install failure & retry behavior by @mintlify[bot] in #4976
  • Wake hosted browser boxes through the control plane, and report presence only while somebody is looking by @chelojimenez in #4915
  • Document hosted browser sleep/wake behavior by @mintlify[bot] in #4977
  • Close the two residual MJ-001 reflection paths in the hosted doctor by @olartgabo in #4884
  • Stop user-supplied eval tag names reaching the chart <style> block by @olartgabo in #4831
  • Enforce the job launcher child-executable invariant where it is enforceable by @olartgabo in #4827
  • Reject a malformed eval id at the v1 boundary instead of paging on it by @olartgabo in #4358
  • Guards that report green without reading a file by @olartgabo in #4979
  • ci: run the PR gate on every pull request, whatever its base by @chelojimenez in #4981
  • User Testing: create-flow validation, a refusal that does not redirect, a locked setup, and a Settings that uses the screen by @ZeHuari in #4955
  • Move the sidebar "New" badge from XAA Debugger to WebMCP by @chelojimenez in #4991
  • Publish MCPJam's UI tools to browser-native WebMCP agents by @chelojimenez in #4987
  • Pull request for mintlify/docs-update-pr-4987-1789260563983 by @mintlify[bot] in #4993
  • Read WebMCP frames with the shared reader, and drop the fallback rungs by @chelojimenez in #4978
  • Send an SDK user-agent by default, and record the caller's by @chelojimenez in #4984
  • Refuse to send a stored on-row credential to a URL it was not saved against (MJ-003, server half) by @olartgabo in #4989
  • Warn before an edit that clears a server's saved credentials (MJ-003, client half) by @olartgabo in #4990
  • Docs: credential-clear warning & destination history by @mintlify[bot] in #4994
  • WebMCP fixes: nine robustness fixes in the browser stack by @chelojimenez in #4988
  • WebMCP fixes: type project secrets into a page without the model seeing them by @chelojimenez in #4996
  • Document browser secret placeholders feature by @mintlify[bot] in #4999
  • Inventory the evaluator-vocabulary renames, and refuse to perform them by @chelojimenez in #4983
  • Pin the evaluator vocabulary, and freeze what a rename may not move by @chelojimenez in #4982
  • Add the canonical evaluator contract beside the score contract by @chelojimenez in #4985
  • fix(servers): honor Auto-connect switch in the connect hook by @saengha in #4925
  • Add assertion(), judge() and runEvaluators, and the assertions subpath by @chelojimenez in #4986
  • Show one Expected Outcome per case, after the last action, marked as the judge's by @chelojimenez in #5005
  • Fix WebMCP browser cleanup hanging after navigation by @chelojimenez in #4995
  • Stamp the MCPJam verification date in npm and desktop builds by @ignaciojimenezr in #5009
  • Stop the Add Server dialog crashing on translated pages by @ignaciojimenezr in #5008
  • Stop the codemod test blocking the vitest worker's RPC channel by @prathmeshpatel in #5012
  • Refresh settings and billing by @prathmeshpatel in #5006
  • What to try: count unchecked items instead of "N left" by @Paoli99 in #4961
  • Show GitHub provenance columns by @ignaciojimenezr in #5002
  • Docs: update GitHub provenance column references by @mintlify[bot] in #5014
  • Grading checks: block unparsable JSON on Save, and start a new check neutral by @Paoli99 in #4974
  • Automatically detect CI metadata for SDK evals by @ignaciojimenezr in #5015
  • Remove stale manual CI metadata section by @mintlify[bot] in #5017
  • Keep the stage rail during blind review, and mask only the judge-decided card by @chelojimenez in #5018
  • Document blind review user-value chain behavior by @mintlify[bot] in #5019
  • Create study: two studies may share a setup; a taken name is what stops you by @ZeHuari in #5016
  • Chat composer: Up and Down walk your own past messages (BB-183) by @ZeHuari in #4950
  • Document message history navigation in chat composer by @mintlify[bot] in #5024
  • User Testing sessions: drop the user value chain panel by @ZeHuari in #5026
  • Say "assertion" and "iteration" in two SDK decision labels by @chelojimenez in #5010
  • feat(evals): persist standard checks and report discovery assertions through UVC by @chelojimenez in #5021
  • Pull request for mintlify/docs-update-pr-5021-1789342611514 by @mintlify[bot] in #5027
  • Keep the CLI badge on a CLI run inside GitHub Actions by @ignaciojimenezr in #5020
  • Add hosted evals GitHub Action by @ignaciojimenezr in #5011
  • Evals: one scorer table for suite settings and case checks by @chelojimenez in #5030
  • Hide unlaunched platform chips behind platform-post-launch by @ignaciojimenezr in #5031
  • Say evaluator, assertion and iteration in the evals UI (Wave 4a) by @chelojimenez in #5007
  • Ship enterprise eval SDK execution, reporting, and evaluator tooling by @chelojimenez in #5022
  • Fix/gh check server display by @ignaciojimenezr in #5034
  • fix(evals): restore Node 20 support and separate reporting cancellation by @chelojimenez in #5037
  • Number the run breadcrumb and heading by @ignaciojimenezr in #5035
  • Evals: group the Add drawer by user-value-chain stage by @chelojimenez in #5038
  • Move client selection into Connect views by @prathmeshpatel in #5039
  • Update client selector docs for Connect view placement by @mintlify[bot] in #5041
  • Use durable eval client identity in inspector and public API by @chelojimenez in #5042
  • Unpublish: document ?scenarioId=, and stop a taken fallback name dead-ending by @olartgabo in #5025
  • SDK: eval suite file dialect 2 (iterations + assertions); resolved shape goes canonical by @chelojimenez in #5044
  • Evals: say why a connection failed, and never blame the server for our refresh by @chelojimenez in #5043
  • Compare runs — first pass (EBB-40) + route 404 (EBB-14) by @chelojimenez in #5053
  • Move run hero measurements into each client-model row by @ignaciojimenezr in #5040
  • SDK: run-knob iterations canonical, repetitions the legacy spelling (S2) by @chelojimenez in #5048
  • Fix/loaded runs count suite scope by @ignaciojimenezr in #5033
  • Bring merged #5045 evaluate changes into main by @ignaciojimenezr in #5061
  • Document run view metrics & case drawer by @mintlify[bot] in #5064
  • User Testing: six fixes from the design review by @ZeHuari in #5062
  • fix: show HTTP status and auth challenge for tool failures by @DaleSeo in #5067
  • Name the org settings entry "Organization keys" by @ignaciojimenezr in #5055
  • Evals: an assertion's policy is Required or Advisory by @chelojimenez in #5065
  • CLI: --iterations canonical, --max-iterations, eval export --schema-version <1|2> by @chelojimenez in #5049
  • v1: read x-mcpjam-eval-vocabulary, and advertise the vocabulary block (A0) by @chelojimenez in #5050
  • v1: vacate the floor — legacyIterations under vocabulary 2 (A1) by @chelojimenez in #5052
  • Evals: evaluator rows disclose progressively; case page drops suite chrome by @chelojimenez in #5051
  • Pull request for mintlify/docs-update-pr-5051-1789422974349 by @mintlify[bot] in #5074
  • Swarms: set iterations per persona, not one number for the whole swarm (BB-194) by @nachocossio in #5054
  • fix(findings): render the summary as one paragraph, not stacked lines (BB-146) by @SebasKoria in #4992
  • Swarms: edit a reused persona's goal, not its card label (UTSC-36) by @nachocossio in #5059
  • findings: a stage finding opens the sessions it is about (BB-146) by @SebasKoria in #4952
  • Changelog: Findings tab stage-based session narrowing by @mintlify[bot] in #5077
  • Revert the MJ-003 server half until the backfill has actually run by @olartgabo in #5029
  • Answer every reason the backend gives for withholding an OAuth token by @olartgabo in #5003
  • Refresh Ding Dong run history, case actions, and Scorecard UI by @ignaciojimenezr in #5063
  • fix(swarms): one "Completed with issues" badge on Overview rows (BB-237) by @SebasKoria in #5081
  • Remove the aggregate goal-completion line graph from Swarms Overview by @SebasKoria in #5078
  • BB-234: warn that Swarms take real actions on the connected server by @SebasKoria in #5075
  • insights: Sankey and clusters only, drop the recommendations rail (BB-230) by @SebasKoria in #5082
  • Fix/evaluator page labels by @ignaciojimenezr in #5085
  • Update Evaluate docs for UI label renames and Overview tab by @mintlify[bot] in #5090
  • feat(evals): publish detailed pull request reports by @ignaciojimenezr in #5046
  • Simplify suite settings into one page with checks by stage by @chelojimenez in #5088
  • Update suite settings UI terminology in evals docs by @mintlify[bot] in #5093
  • Infer SDK reporting server names and fix missing run labels by @ignaciojimenezr in #5092
  • Update serverNames auto-inference docs by @mintlify[bot] in #5095
  • Open test definitions when clicking names in run results by @ignaciojimenezr in #5096
  • Document test name click behavior in run view by @mintlify[bot] in #5098
  • Fix env-backed eval runs on the local inspector by @ignaciojimenezr in #5100
  • Polish Evaluate charts, run headers, and comparison tables by @ignaciojimenezr in #5099
  • Update run view docs: HOLD/SHIP pills and deltas timing by @mintlify[bot] in #5101
  • Narrow eval filter options using other active selections by @ignaciojimenezr in #5105
  • Show the credits wall when the agent panel or case generation hits the limit by @ignaciojimenezr in #5106
  • Run Supervisor (inspector): execution budgets, a clock per turn, and swarm watchdogs by @chelojimenez in #5091
  • Pull request for mintlify/docs-update-pr-5091-1789464929985 by @mintlify[bot] in #5109
  • Evals: one grading policy, one vocabulary (G0a, G0c, G1, G2, G3) by @chelojimenez in #5070
  • User Testing: say Team members everywhere, not Project members (BB-203) by @ZeHuari in #4850
  • Sessions: reuse the Playground's tool card and JSON viewer by @ZeHuari in #5097
  • Changelog: session viewer UI improvements by @mintlify[bot] in #5114
  • fix(sessions): match Playground's transcript — no generic bubble, coloured JSON (BB-239) by @SebasKoria in #5080
  • Changelog: session transcript visual parity by @mintlify[bot] in #5115
  • Fence em dashes out of new client copy (BB-240) by @nachocossio in #5089
  • One picker: servers and server groups as sibling tabs by @weimar-torres-herrera in #4695
  • Swarm Insights: pin Clusters to the pane, unclip the cluster rail by @nachocossio in #5116
  • Improve V2 pricing and credit refill management by @prathmeshpatel in #5104
  • Remove baseline comparison from the eval suite settings UI by @ignaciojimenezr in #5110
  • Update quality gate settings docs for baseline UI removal by @mintlify[bot] in #5120
  • Remove the Suggested assertions section from the trial scorecard by @ignaciojimenezr in #5123
  • swarms: take a typed iteration count instead of appending it (BB-246) by @nachocossio in #5121
  • swarms: restore the header headline on a shared landing shell (BB-236) by @nachocossio in #5117
  • Run Supervisor follow-up: make a fired clock decide, and close the escapes it opened by @chelojimenez in #5111
  • Evaluate: findings panel and the empty-step engine fix (lean rework of #5066) by @chelojimenez in #5112
  • Document run page findings panel and scorecard updates by @mintlify[bot] in #5126
  • Playground: Tailwind never scanned widget-react, so fullscreen Apps lost their z-index by @olartgabo in #5086
  • changeset: drop @mcpjam/design-system from scroll-fade-utility by @chelojimenez in #5128
  • Release: wait for the registry to serve a first-published evaluators before changesets looks it up by @chelojimenez in #5130

New Contributors

Full Changelog: v3.5.4...v3.6.0

Don't miss a new inspector release

NewReleases is sending notifications on new releases.