@langwatch/mcp-server 2.0.0 → 2.2.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (430) hide show
  1. package/.gitignore +7 -0
  2. package/.npmignore +41 -0
  3. package/CHANGELOG.md +280 -200
  4. package/CONTRIBUTING.md +10 -0
  5. package/README.md +82 -52
  6. package/dist/add-dashboard-widget-4ZC5UN73.js +50 -0
  7. package/dist/add-dashboard-widget-ACC6QYBG.js +49 -0
  8. package/dist/archive-project-KWFM3I5Z.js +19 -0
  9. package/dist/{archive-project-4EQGLBJJ.js → archive-project-SPVHXK6A.js} +4 -4
  10. package/dist/archive-run-plan-65QN3XRX.js +16 -0
  11. package/dist/archive-run-plan-Z4SYE6BP.js +15 -0
  12. package/dist/{archive-scenario-W6BXYDZ3.js → archive-scenario-A6QV5LOW.js} +4 -4
  13. package/dist/archive-scenario-FXEYQNEH.js +19 -0
  14. package/dist/archive-test-suite-5ADQ3KPQ.js +15 -0
  15. package/dist/archive-test-suite-NKZDZY3D.js +16 -0
  16. package/dist/assign-prompt-tag-5IUZXJ2T.js +23 -0
  17. package/dist/{assign-prompt-tag-VLXUWEKU.js → assign-prompt-tag-HBZLBIFD.js} +8 -4
  18. package/dist/{chunk-CDYQOY5U.js → chunk-3OCG3QPU.js} +7 -10
  19. package/dist/chunk-3YLN3ATJ.js +241 -0
  20. package/dist/chunk-5FNCQRCX.js +91 -0
  21. package/dist/chunk-6YZUOD34.js +55 -0
  22. package/dist/{chunk-WVI5CJ6S.js → chunk-7AQLPVFL.js} +2 -2
  23. package/dist/chunk-7VNSRCXO.js +129 -0
  24. package/dist/{chunk-7RHG56OH.js → chunk-B4J3PJLX.js} +11 -10
  25. package/dist/{chunk-ZXKLPC2E.js → chunk-BIIKE3EI.js} +1 -1
  26. package/dist/{chunk-37SMPMQS.js → chunk-DMM3JQ42.js} +7 -2
  27. package/dist/{chunk-NUFPTHKU.js → chunk-FOKVY3BK.js} +1 -1
  28. package/dist/{chunk-WT4PO4A3.js → chunk-H6WGMG2T.js} +10 -7
  29. package/dist/{chunk-TOXJQDIZ.js → chunk-HGELNIVY.js} +135 -51
  30. package/dist/{chunk-UJPXFA26.js → chunk-HHMNZAPD.js} +101 -74
  31. package/dist/{chunk-J42FWRJ2.js → chunk-I7ATIZED.js} +7 -7
  32. package/dist/{chunk-AQJO2EI5.js → chunk-ILZDO25L.js} +7 -10
  33. package/dist/{chunk-4QC7SA2L.js → chunk-J7PQKBLD.js} +884 -559
  34. package/dist/{chunk-YX3A3CFM.js → chunk-JRJPH75H.js} +7 -13
  35. package/dist/chunk-K2I7Z3LZ.js +28 -0
  36. package/dist/{chunk-5CCZFUXS.js → chunk-NKMTIHIJ.js} +6 -9
  37. package/dist/chunk-O7EMWEVJ.js +44 -0
  38. package/dist/chunk-PTDR56Z3.js +37 -0
  39. package/dist/chunk-RABYBFX5.js +43 -0
  40. package/dist/chunk-SUPUUC4W.js +64 -0
  41. package/dist/{chunk-SFB4MU6H.js → chunk-SXFC3FNH.js} +4 -4
  42. package/dist/chunk-U24EACC4.js +67 -0
  43. package/dist/chunk-UNJM3BIL.js +21 -0
  44. package/dist/{chunk-ZWIL4EFL.js → chunk-UTJCWXPZ.js} +4 -4
  45. package/dist/chunk-UWLE5T2F.js +312 -0
  46. package/dist/chunk-VMWJD732.js +57 -0
  47. package/dist/chunk-XUIJYNRJ.js +15 -0
  48. package/dist/chunk-Y34DHR5K.js +67 -0
  49. package/dist/chunk-YCARO42Z.js +516 -0
  50. package/dist/config-EAKW6F26.js +16 -0
  51. package/dist/config.cjs +109 -0
  52. package/dist/config.js +5 -4
  53. package/dist/{create-agent-7XHO6UIS.js → create-agent-AYZ3LB3V.js} +4 -4
  54. package/dist/create-agent-S5ZBCEFT.js +23 -0
  55. package/dist/create-annotation-LHNCLHYP.js +29 -0
  56. package/dist/{create-annotation-YQXRQC6A.js → create-annotation-VUSXULQK.js} +10 -5
  57. package/dist/{create-api-key-YUS36KBU.js → create-api-key-NNYKARMG.js} +5 -7
  58. package/dist/create-api-key-T4KSNPKB.js +24 -0
  59. package/dist/create-dashboard-DHVFMEU4.js +18 -0
  60. package/dist/{create-dashboard-4Z5JO5LB.js → create-dashboard-WTKKS5Z6.js} +4 -4
  61. package/dist/{create-dataset-E6NMSVX7.js → create-dataset-2MC4SD4O.js} +6 -6
  62. package/dist/create-dataset-LUP5PVDM.js +22 -0
  63. package/dist/create-dataset-records-NSKBZP3G.js +20 -0
  64. package/dist/{create-dataset-records-SKHPRD2K.js → create-dataset-records-W4GLN2FV.js} +5 -7
  65. package/dist/{create-evaluator-JXF5PF33.js → create-evaluator-643G6LXL.js} +4 -4
  66. package/dist/create-evaluator-HMTYO5SS.js +27 -0
  67. package/dist/create-mcp-server.cjs +9513 -0
  68. package/dist/create-mcp-server.js +2647 -5
  69. package/dist/create-project-6N7PIYRY.js +44 -0
  70. package/dist/{create-project-CS2CB57Y.js → create-project-KAIC46UY.js} +4 -4
  71. package/dist/{create-prompt-F5RMX3L3.js → create-prompt-E67LXIYT.js} +5 -7
  72. package/dist/create-prompt-PKJ7EDRR.js +37 -0
  73. package/dist/create-prompt-tag-4YO3FSJ3.js +17 -0
  74. package/dist/{create-prompt-tag-GMRWTEFR.js → create-prompt-tag-OPQUEXWB.js} +3 -3
  75. package/dist/create-scenario-5PJBMAXO.js +33 -0
  76. package/dist/{create-scenario-JDR2VB2H.js → create-scenario-UEM77OE7.js} +11 -4
  77. package/dist/create-test-suite-4IPXVSFQ.js +37 -0
  78. package/dist/create-test-suite-SZEPVXSE.js +36 -0
  79. package/dist/create-trigger-5HZSCRAN.js +49 -0
  80. package/dist/create-trigger-PRELMT6J.js +48 -0
  81. package/dist/delete-agent-JUIZWAXG.js +18 -0
  82. package/dist/{delete-agent-O3EEH4S5.js → delete-agent-QMEK3XHK.js} +4 -4
  83. package/dist/delete-annotation-HY3O2OGR.js +17 -0
  84. package/dist/{delete-annotation-XP6FQM7A.js → delete-annotation-VKIQWUYW.js} +4 -4
  85. package/dist/delete-dashboard-76N66XXL.js +18 -0
  86. package/dist/{delete-dashboard-RFUNYCNP.js → delete-dashboard-JVIVHNJM.js} +4 -4
  87. package/dist/delete-dataset-HQCR66MG.js +18 -0
  88. package/dist/{delete-dataset-3MJFRDLP.js → delete-dataset-WKRUW4FP.js} +5 -7
  89. package/dist/{delete-dataset-records-WPMU5GL5.js → delete-dataset-records-C2IQFFVQ.js} +5 -7
  90. package/dist/delete-dataset-records-SDUHQLLM.js +19 -0
  91. package/dist/delete-evaluator-4SNSCLEE.js +16 -0
  92. package/dist/delete-evaluator-QP3QHTLN.js +15 -0
  93. package/dist/delete-prompt-tag-NZ4JQMYG.js +17 -0
  94. package/dist/{delete-prompt-tag-KNMZVSPB.js → delete-prompt-tag-YYQ4VMAD.js} +3 -3
  95. package/dist/{delete-workflow-3NUXO46U.js → delete-workflow-5J46T73T.js} +4 -4
  96. package/dist/delete-workflow-JSWABL5F.js +17 -0
  97. package/dist/discover-evaluator-schema-UHWA5HVL.js +120 -0
  98. package/dist/discover-evaluator-schema-XKCLLCB3.js +118 -0
  99. package/dist/discover-scenario-schema-NOJY6S7E.js +137 -0
  100. package/dist/discover-scenario-schema-OOFIQKOG.js +139 -0
  101. package/dist/{discover-schema-L6QDPON3.js → discover-schema-55YJBFGQ.js} +181 -186
  102. package/dist/discover-schema-A33SLMXW.js +442 -0
  103. package/dist/get-agent-LJQX7OF4.js +69 -0
  104. package/dist/get-agent-OTIYUFZA.js +70 -0
  105. package/dist/get-analytics-DCS72353.js +80 -0
  106. package/dist/get-analytics-IQZKJHLW.js +79 -0
  107. package/dist/get-dashboard-5NYJEIX2.js +29 -0
  108. package/dist/{get-dashboard-KHBEGZGL.js → get-dashboard-QMTYDQAZ.js} +4 -4
  109. package/dist/get-dataset-ALSP7RTX.js +64 -0
  110. package/dist/get-dataset-X4B5JDG2.js +63 -0
  111. package/dist/{get-evaluator-ADMKVWYZ.js → get-evaluator-ISMC4LLI.js} +4 -4
  112. package/dist/get-evaluator-XEUZGZCH.js +47 -0
  113. package/dist/{get-project-RAK7URQJ.js → get-project-5VKROSDH.js} +4 -4
  114. package/dist/get-project-RM6FRCBK.js +26 -0
  115. package/dist/get-prompt-OWXJZHQQ.js +95 -0
  116. package/dist/get-prompt-W52DY36E.js +94 -0
  117. package/dist/get-run-plan-BE3M2AUX.js +59 -0
  118. package/dist/get-run-plan-RUXQ6PAX.js +58 -0
  119. package/dist/get-scenario-RNM7BAX2.js +37 -0
  120. package/dist/{get-scenario-ZTAJJL5T.js → get-scenario-YM4ODWW6.js} +8 -4
  121. package/dist/get-simulation-run-6GIATQE5.js +69 -0
  122. package/dist/{get-simulation-run-T2ZQWYH2.js → get-simulation-run-LCFSMPGQ.js} +33 -20
  123. package/dist/get-test-suite-GC2O7YH6.js +49 -0
  124. package/dist/get-test-suite-P5LJ6VLF.js +48 -0
  125. package/dist/get-trace-KBRVV4NN.js +45 -0
  126. package/dist/{get-trace-K7U234Q4.js → get-trace-SEF3FMR5.js} +11 -13
  127. package/dist/get-workflow-7PVED7JE.js +25 -0
  128. package/dist/{get-workflow-L5L2PSK3.js → get-workflow-YXULW4GC.js} +4 -4
  129. package/dist/{http-security-EU3CHTY7.js → http-security-DNLBNQZL.js} +2 -2
  130. package/dist/{http-server-YXHFPPGA.js → http-server-IR4QOS43.js} +4071 -342
  131. package/dist/index.js +60221 -38077
  132. package/dist/langwatch-api-annotations-6SA4OB4C.js +17 -0
  133. package/dist/{langwatch-api-annotations-FDAS6NQ7.js → langwatch-api-annotations-TBHLCQ5R.js} +4 -4
  134. package/dist/langwatch-api-dashboards-6EO2VVZW.js +17 -0
  135. package/dist/{langwatch-api-dashboards-JTVCS7ZJ.js → langwatch-api-dashboards-NHU5BZD7.js} +4 -4
  136. package/dist/langwatch-api-monitors-CBF3FGIM.js +47 -0
  137. package/dist/{langwatch-api-monitors-3J7UIQME.js → langwatch-api-monitors-YNWA2M5Y.js} +11 -16
  138. package/dist/langwatch-api-secrets-ADTOJELP.js +56 -0
  139. package/dist/langwatch-api-secrets-RNFO4LF2.js +55 -0
  140. package/dist/langwatch-api-triggers-KQDQC6L5.js +22 -0
  141. package/dist/langwatch-api-triggers-QS7HDIVW.js +23 -0
  142. package/dist/{langwatch-api-workflows-NBJX5KOG.js → langwatch-api-workflows-OKC4FXIF.js} +4 -4
  143. package/dist/langwatch-api-workflows-XGJE6SR5.js +15 -0
  144. package/dist/list-agents-HVYWEQMI.js +43 -0
  145. package/dist/list-agents-RBAOCF53.js +42 -0
  146. package/dist/list-annotations-HJH4NBLD.js +38 -0
  147. package/dist/{list-annotations-ZTVP4EEL.js → list-annotations-J4UAHSUN.js} +10 -5
  148. package/dist/{list-api-keys-CWD4UQT5.js → list-api-keys-7UTJRZDG.js} +10 -5
  149. package/dist/list-api-keys-A2PMV435.js +42 -0
  150. package/dist/list-dashboard-boards-OPWROHMG.js +33 -0
  151. package/dist/list-dashboard-boards-U2GJ7GC5.js +34 -0
  152. package/dist/{list-dashboards-PWG6P6KR.js → list-dashboards-ROSYFLDQ.js} +4 -4
  153. package/dist/list-dashboards-RQ635ZU7.js +32 -0
  154. package/dist/{list-dataset-records-IOE7LQXS.js → list-dataset-records-M5TKA6LK.js} +14 -10
  155. package/dist/list-dataset-records-WFM7F33C.js +43 -0
  156. package/dist/{list-datasets-ZT4JYUDT.js → list-datasets-B5X4P7BQ.js} +9 -13
  157. package/dist/list-datasets-THUKQHPX.js +56 -0
  158. package/dist/{list-evaluators-SJHC7VBH.js → list-evaluators-MYH2OVCI.js} +5 -7
  159. package/dist/list-evaluators-NLVWF6K7.js +32 -0
  160. package/dist/{list-model-providers-PPD2WPD2.js → list-model-providers-3JUGP4YI.js} +4 -4
  161. package/dist/list-model-providers-H3XKORFO.js +35 -0
  162. package/dist/{list-projects-K7QAIX2I.js → list-projects-6EJ5YYEW.js} +5 -7
  163. package/dist/list-projects-YNNSRMLC.js +40 -0
  164. package/dist/{list-prompt-tags-FOO6TNHB.js → list-prompt-tags-NVFQO6WX.js} +3 -3
  165. package/dist/list-prompt-tags-XUQTTAH4.js +23 -0
  166. package/dist/{list-prompts-SC2HAUFG.js → list-prompts-HSIMMHTF.js} +4 -6
  167. package/dist/list-prompts-IRI6AIXO.js +30 -0
  168. package/dist/list-run-plans-PX2XI6EK.js +46 -0
  169. package/dist/list-run-plans-QALLCN2K.js +45 -0
  170. package/dist/list-scenarios-QZRPVHS4.js +46 -0
  171. package/dist/list-scenarios-SSJQEMJP.js +45 -0
  172. package/dist/list-simulation-runs-KNOLGMDV.js +53 -0
  173. package/dist/{list-simulation-runs-KF5THG5H.js → list-simulation-runs-OEADLLQR.js} +19 -9
  174. package/dist/list-test-suites-ANAQBLDG.js +34 -0
  175. package/dist/list-test-suites-N4WVLBKK.js +35 -0
  176. package/dist/list-trigger-fires-JV3WFYYG.js +19 -0
  177. package/dist/list-trigger-fires-WTW4DVIW.js +20 -0
  178. package/dist/list-triggers-SQNIG35Q.js +45 -0
  179. package/dist/list-triggers-V4Y356AK.js +46 -0
  180. package/dist/{list-workflows-QSJWURCK.js → list-workflows-IMSX7SXU.js} +5 -7
  181. package/dist/list-workflows-ZHTX5LCL.js +33 -0
  182. package/dist/{rename-prompt-tag-EKWU5QF3.js → rename-prompt-tag-IQI7MHOI.js} +3 -3
  183. package/dist/rename-prompt-tag-PA2BRIE6.js +18 -0
  184. package/dist/rename-test-suite-4J7UX5S2.js +21 -0
  185. package/dist/rename-test-suite-NNVLMLKY.js +20 -0
  186. package/dist/report-issue-AS34ZMJF.js +898 -0
  187. package/dist/report-issue-DCZFLGDO.js +899 -0
  188. package/dist/rerun-run-plan-TG2V4WM4.js +22 -0
  189. package/dist/rerun-run-plan-ZI5OY3VA.js +21 -0
  190. package/dist/revoke-api-key-3J3ZA4UB.js +15 -0
  191. package/dist/{revoke-api-key-FTRTZMOF.js → revoke-api-key-SFMEFWJC.js} +4 -4
  192. package/dist/run-agent-W4VJP72Z.js +75 -0
  193. package/dist/run-agent-YNV6ZHW4.js +74 -0
  194. package/dist/run-plan-6C77GWG4.js +38 -0
  195. package/dist/run-plan-PAARQSI4.js +39 -0
  196. package/dist/run-query-KJDH544C.js +47 -0
  197. package/dist/run-query-T4QW2Z5O.js +46 -0
  198. package/dist/run-test-suite-SQJELEZX.js +26 -0
  199. package/dist/run-test-suite-STB6NUJM.js +25 -0
  200. package/dist/{run-workflow-QPXC627O.js → run-workflow-D6PQNBI6.js} +4 -4
  201. package/dist/{run-agent-6SM4XNA7.js → run-workflow-SP32SO5M.js} +10 -11
  202. package/dist/search-traces-HCBGOG6R.js +170 -0
  203. package/dist/search-traces-HSTQ6WTQ.js +171 -0
  204. package/dist/{set-model-provider-3WKT63ZN.js → set-model-provider-GF7NXUJA.js} +4 -4
  205. package/dist/set-model-provider-KDQ7C4EZ.js +33 -0
  206. package/dist/{update-agent-SMAEHLXH.js → update-agent-DWG43T2P.js} +4 -4
  207. package/dist/update-agent-MXNOBS4H.js +20 -0
  208. package/dist/update-dashboard-widget-CJKJLCDY.js +71 -0
  209. package/dist/update-dashboard-widget-CYQDS2NR.js +70 -0
  210. package/dist/update-dataset-DQH3OHOU.js +22 -0
  211. package/dist/{update-dataset-BDUK7TUO.js → update-dataset-SQZ6T4WX.js} +6 -6
  212. package/dist/{update-dataset-record-4ZSDRBSW.js → update-dataset-record-3AT4IPBC.js} +4 -4
  213. package/dist/update-dataset-record-7PLDGTMR.js +19 -0
  214. package/dist/update-evaluator-4LFTKFPT.js +24 -0
  215. package/dist/{update-evaluator-LS6TIBY7.js → update-evaluator-IGTX6PX4.js} +4 -4
  216. package/dist/{update-project-LKVYXENH.js → update-project-PLJIOZS6.js} +4 -4
  217. package/dist/update-project-QVHSJRUP.js +24 -0
  218. package/dist/update-prompt-GDKRYJI3.js +140 -0
  219. package/dist/update-prompt-MY263DYF.js +141 -0
  220. package/dist/update-scenario-7KA3JDXA.js +34 -0
  221. package/dist/{update-scenario-NEXCAQI7.js → update-scenario-MOHRQKAQ.js} +11 -4
  222. package/package.json +39 -44
  223. package/pyproject.toml +8 -8
  224. package/src/config.d.ts +4 -0
  225. package/src/config.ts +20 -29
  226. package/src/create-mcp-server.ts +1021 -596
  227. package/src/documentation-fetch.ts +21 -20
  228. package/src/http-security.ts +166 -135
  229. package/src/http-server.ts +189 -224
  230. package/src/index.ts +14 -16
  231. package/src/langwatch-api-agents.ts +202 -21
  232. package/src/langwatch-api-annotations.ts +11 -12
  233. package/src/langwatch-api-api-keys.ts +8 -9
  234. package/src/langwatch-api-boards.ts +46 -0
  235. package/src/langwatch-api-dashboard-widgets.ts +112 -0
  236. package/src/langwatch-api-dashboards.ts +13 -10
  237. package/src/langwatch-api-datasets.ts +98 -32
  238. package/src/langwatch-api-evaluators.ts +13 -19
  239. package/src/langwatch-api-model-providers.ts +4 -4
  240. package/src/langwatch-api-monitors.ts +14 -18
  241. package/src/langwatch-api-projects.ts +27 -9
  242. package/src/langwatch-api-query.ts +114 -0
  243. package/src/langwatch-api-run-plans.ts +140 -0
  244. package/src/langwatch-api-scenarios.ts +17 -10
  245. package/src/langwatch-api-secrets.ts +31 -24
  246. package/src/langwatch-api-simulation-runs.ts +10 -7
  247. package/src/langwatch-api-test-suites.ts +116 -0
  248. package/src/langwatch-api-triggers.ts +96 -34
  249. package/src/langwatch-api-workflows.ts +9 -9
  250. package/src/langwatch-api.ts +143 -109
  251. package/src/public-http-request.ts +21 -12
  252. package/src/schemas/analytics-metrics.ts +3 -15
  253. package/src/schemas/create-dataset.ts +3 -4
  254. package/src/schemas/run-plan.ts +76 -0
  255. package/src/schemas/suite-fields.ts +132 -0
  256. package/src/schemas/triggers.ts +365 -0
  257. package/src/tools/add-dashboard-widget.ts +63 -0
  258. package/src/tools/archive-project.ts +2 -4
  259. package/src/tools/archive-run-plan.ts +10 -0
  260. package/src/tools/archive-scenario.ts +2 -4
  261. package/src/tools/archive-test-suite.ts +10 -0
  262. package/src/tools/assign-prompt-tag.ts +6 -2
  263. package/src/tools/create-agent.ts +1 -1
  264. package/src/tools/create-annotation.ts +7 -2
  265. package/src/tools/create-api-key.ts +4 -6
  266. package/src/tools/create-dashboard.ts +1 -1
  267. package/src/tools/create-dataset-records.ts +4 -8
  268. package/src/tools/create-dataset.ts +5 -7
  269. package/src/tools/create-evaluator.ts +3 -5
  270. package/src/tools/create-project.ts +2 -2
  271. package/src/tools/create-prompt-tag.ts +2 -4
  272. package/src/tools/create-prompt.ts +9 -18
  273. package/src/tools/create-scenario.ts +11 -5
  274. package/src/tools/create-test-suite.ts +38 -0
  275. package/src/tools/create-trigger.ts +55 -0
  276. package/src/tools/dashboard-exists.ts +25 -0
  277. package/src/tools/dashboard-widget-queries.ts +32 -0
  278. package/src/tools/delete-agent.ts +1 -1
  279. package/src/tools/delete-annotation.ts +1 -1
  280. package/src/tools/delete-dashboard.ts +1 -1
  281. package/src/tools/delete-dataset-records.ts +4 -8
  282. package/src/tools/delete-dataset.ts +3 -7
  283. package/src/tools/delete-evaluator.ts +4 -6
  284. package/src/tools/delete-prompt-tag.ts +2 -4
  285. package/src/tools/delete-workflow.ts +1 -1
  286. package/src/tools/discover-evaluator-schema.ts +34 -25
  287. package/src/tools/discover-scenario-schema.ts +86 -12
  288. package/src/tools/discover-schema.ts +149 -27
  289. package/src/tools/experiment-run-status.ts +2 -8
  290. package/src/tools/format-dataset-mutation.ts +5 -10
  291. package/src/tools/format-run-plan.ts +54 -0
  292. package/src/tools/format-scenario.ts +12 -0
  293. package/src/tools/format-suite-details.ts +90 -0
  294. package/src/tools/get-agent.ts +54 -3
  295. package/src/tools/get-analytics.ts +52 -46
  296. package/src/tools/get-dashboard.ts +1 -1
  297. package/src/tools/get-dataset.ts +34 -16
  298. package/src/tools/get-evaluator.ts +4 -11
  299. package/src/tools/get-experiment-results.ts +219 -186
  300. package/src/tools/get-project.ts +2 -4
  301. package/src/tools/get-prompt.ts +100 -37
  302. package/src/tools/get-run-plan.ts +60 -0
  303. package/src/tools/get-scenario.ts +6 -5
  304. package/src/tools/get-simulation-run.ts +32 -20
  305. package/src/tools/get-test-suite.ts +46 -0
  306. package/src/tools/get-trace.ts +16 -17
  307. package/src/tools/get-workflow.ts +1 -1
  308. package/src/tools/list-agents.ts +20 -3
  309. package/src/tools/list-annotations.ts +8 -5
  310. package/src/tools/list-api-keys.ts +7 -2
  311. package/src/tools/list-dashboard-boards.ts +28 -0
  312. package/src/tools/list-dashboards.ts +1 -1
  313. package/src/tools/list-dataset-records.ts +8 -6
  314. package/src/tools/list-datasets.ts +12 -23
  315. package/src/tools/list-evaluators.ts +4 -8
  316. package/src/tools/list-experiment-runs.ts +5 -9
  317. package/src/tools/list-experiments.ts +5 -11
  318. package/src/tools/list-model-providers.ts +3 -5
  319. package/src/tools/list-projects.ts +2 -4
  320. package/src/tools/list-prompt-tags.ts +3 -3
  321. package/src/tools/list-prompts.ts +4 -8
  322. package/src/tools/list-run-plans.ts +47 -0
  323. package/src/tools/list-scenarios.ts +26 -17
  324. package/src/tools/list-simulation-runs.ts +13 -5
  325. package/src/tools/list-test-suites.ts +35 -0
  326. package/src/tools/list-trigger-fires.ts +17 -0
  327. package/src/tools/list-triggers.ts +39 -0
  328. package/src/tools/list-workflows.ts +2 -4
  329. package/src/tools/rename-prompt-tag.ts +1 -1
  330. package/src/tools/rename-test-suite.ts +15 -0
  331. package/src/tools/report-issue.ts +5 -9
  332. package/src/tools/rerun-run-plan.ts +18 -0
  333. package/src/tools/revoke-api-key.ts +2 -4
  334. package/src/tools/run-agent.ts +65 -14
  335. package/src/tools/run-experiment.ts +36 -37
  336. package/src/tools/run-plan.ts +46 -0
  337. package/src/tools/run-query.ts +47 -0
  338. package/src/tools/run-test-suite.ts +27 -0
  339. package/src/tools/run-workflow.ts +2 -5
  340. package/src/tools/search-traces.ts +161 -47
  341. package/src/tools/set-model-provider.ts +4 -6
  342. package/src/tools/test-agent.ts +18 -0
  343. package/src/tools/update-agent.ts +1 -1
  344. package/src/tools/update-dashboard-widget.ts +97 -0
  345. package/src/tools/update-dataset-record.ts +1 -1
  346. package/src/tools/update-dataset.ts +5 -7
  347. package/src/tools/update-evaluator.ts +3 -5
  348. package/src/tools/update-project.ts +1 -1
  349. package/src/tools/update-prompt.ts +185 -15
  350. package/src/tools/update-scenario.ts +11 -5
  351. package/src/tools/update-test-suite.ts +40 -0
  352. package/src/utils/date-parsing.ts +1 -1
  353. package/src/utils/escape-markdown.ts +3 -8
  354. package/src/utils/format-evaluations.ts +6 -6
  355. package/src/utils/markdown-table.ts +58 -0
  356. package/src/utils/trace-id-shape.ts +20 -0
  357. package/tsconfig.app.json +24 -0
  358. package/tsconfig.json +19 -12
  359. package/uv.lock +416 -296
  360. package/vitest.integration.config.js +10 -0
  361. package/vitest.scenario.config.js +9 -0
  362. package/.eslintrc.cjs +0 -36
  363. package/dist/archive-suite-4T6BSRS7.js +0 -16
  364. package/dist/chunk-77BNWQ3U.js +0 -86
  365. package/dist/chunk-A2LEVYQ5.js +0 -217
  366. package/dist/chunk-TJCXFNDT.js +0 -24
  367. package/dist/chunk-YZLYHLLI.js +0 -48
  368. package/dist/config.d.ts +0 -17
  369. package/dist/create-mcp-server.d.ts +0 -10
  370. package/dist/create-suite-LZS6QIAF.js +0 -39
  371. package/dist/delete-evaluator-33SP2JFX.js +0 -16
  372. package/dist/discover-evaluator-schema-XTAD4X7D.js +0 -2191
  373. package/dist/discover-scenario-schema-FE64QK2D.js +0 -67
  374. package/dist/get-agent-Y7MWZ65F.js +0 -29
  375. package/dist/get-analytics-TETCSOK6.js +0 -73
  376. package/dist/get-dataset-RCMU5VAH.js +0 -49
  377. package/dist/get-prompt-OY5B7TRI.js +0 -46
  378. package/dist/get-suite-J7JVWO6O.js +0 -45
  379. package/dist/index.d.ts +0 -2
  380. package/dist/langwatch-api-secrets-BUBSNCNO.js +0 -41
  381. package/dist/langwatch-api-triggers-EDFUCJN6.js +0 -31
  382. package/dist/list-agents-KFHKGPXV.js +0 -33
  383. package/dist/list-scenarios-F2WYNRIE.js +0 -41
  384. package/dist/list-suites-CRY4VWW6.js +0 -43
  385. package/dist/report-issue-XYVAPAYJ.js +0 -412
  386. package/dist/run-suite-NKZCJ2IN.js +0 -30
  387. package/dist/search-traces-A6HFKPLJ.js +0 -79
  388. package/dist/update-prompt-ODCRQRN6.js +0 -25
  389. package/dist/update-suite-SNO3GS7T.js +0 -40
  390. package/src/__tests__/all-tools.integration.test.ts +0 -1626
  391. package/src/__tests__/config.unit.test.ts +0 -202
  392. package/src/__tests__/datasets.integration.test.ts +0 -620
  393. package/src/__tests__/datasets.unit.test.ts +0 -244
  394. package/src/__tests__/date-parsing.unit.test.ts +0 -78
  395. package/src/__tests__/discover-evaluator-schema.unit.test.ts +0 -89
  396. package/src/__tests__/discover-schema.unit.test.ts +0 -118
  397. package/src/__tests__/docker.integration.test.ts +0 -318
  398. package/src/__tests__/documentation-fetch-security.integration.test.ts +0 -182
  399. package/src/__tests__/evaluator-tools.unit.test.ts +0 -262
  400. package/src/__tests__/experiment-list-tools.unit.test.ts +0 -157
  401. package/src/__tests__/experiment-status.unit.test.ts +0 -194
  402. package/src/__tests__/feature-map-tool-drift.unit.test.ts +0 -93
  403. package/src/__tests__/get-experiment-results.unit.test.ts +0 -313
  404. package/src/__tests__/http-security-origin.unit.test.ts +0 -97
  405. package/src/__tests__/http-security-sessions.unit.test.ts +0 -209
  406. package/src/__tests__/http-security-verifier.unit.test.ts +0 -248
  407. package/src/__tests__/http-server-allocation.integration.test.ts +0 -279
  408. package/src/__tests__/http-server-origin.integration.test.ts +0 -169
  409. package/src/__tests__/http-server-session-auth.integration.test.ts +0 -341
  410. package/src/__tests__/http-transport.integration.test.ts +0 -395
  411. package/src/__tests__/integration.integration.test.ts +0 -288
  412. package/src/__tests__/langwatch-api.unit.test.ts +0 -661
  413. package/src/__tests__/model-provider-tools.unit.test.ts +0 -190
  414. package/src/__tests__/project-api-key-tools.unit.test.ts +0 -451
  415. package/src/__tests__/public-http-request-security.integration.test.ts +0 -118
  416. package/src/__tests__/report-issue.integration.test.ts +0 -147
  417. package/src/__tests__/scenario-tools.integration.test.ts +0 -286
  418. package/src/__tests__/scenario-tools.unit.test.ts +0 -188
  419. package/src/__tests__/schemas.unit.test.ts +0 -85
  420. package/src/__tests__/support/http-server-harness.ts +0 -138
  421. package/src/__tests__/tools.unit.test.ts +0 -924
  422. package/src/langwatch-api-suites.ts +0 -118
  423. package/src/schemas/filter-fields.ts +0 -119
  424. package/src/schemas/index.ts +0 -4
  425. package/src/tools/archive-suite.ts +0 -12
  426. package/src/tools/create-suite.ts +0 -42
  427. package/src/tools/get-suite.ts +0 -47
  428. package/src/tools/list-suites.ts +0 -43
  429. package/src/tools/run-suite.ts +0 -28
  430. package/src/tools/update-suite.ts +0 -44
@@ -1,661 +0,0 @@
1
- import { describe, it, expect, beforeEach, afterEach, vi } from "vitest";
2
- import { initConfig } from "../config.js";
3
-
4
- const TEST_ENDPOINT = "https://test.langwatch.ai";
5
- const TEST_API_KEY = "test-key";
6
-
7
- describe("langwatch-api", () => {
8
- let mockFetch: ReturnType<typeof vi.fn>;
9
-
10
- beforeEach(() => {
11
- initConfig({ apiKey: TEST_API_KEY, endpoint: TEST_ENDPOINT });
12
- mockFetch = vi.fn();
13
- vi.stubGlobal("fetch", mockFetch);
14
- });
15
-
16
- afterEach(() => {
17
- vi.unstubAllGlobals();
18
- });
19
-
20
- function mockJsonResponse(data: unknown, status = 200) {
21
- mockFetch.mockResolvedValueOnce({
22
- ok: status >= 200 && status < 300,
23
- status,
24
- text: async () => JSON.stringify(data),
25
- json: async () => data,
26
- });
27
- }
28
-
29
- function mockErrorResponse(status: number, body: string) {
30
- mockFetch.mockResolvedValueOnce({
31
- ok: false,
32
- status,
33
- text: async () => body,
34
- });
35
- }
36
-
37
- function mock204Response() {
38
- mockFetch.mockResolvedValueOnce({
39
- ok: true,
40
- status: 204,
41
- text: async () => "",
42
- json: async () => { throw new Error("No content"); },
43
- });
44
- }
45
-
46
- describe("searchTraces()", () => {
47
- it("sends POST to /api/traces/search with format digest by default", async () => {
48
- const { searchTraces } = await import("../langwatch-api.js");
49
- const responseData = { traces: [] };
50
- mockJsonResponse(responseData);
51
-
52
- const result = await searchTraces({
53
- startDate: 1000,
54
- endDate: 2000,
55
- query: "hello",
56
- });
57
-
58
- const [calledUrl, calledOptions] = mockFetch.mock.calls[0]!;
59
- expect(calledUrl).toBe(`${TEST_ENDPOINT}/api/traces/search`);
60
- expect(calledOptions.method).toBe("POST");
61
- expect(calledOptions.headers["X-Auth-Token"]).toBe(TEST_API_KEY);
62
- expect(calledOptions.headers["Content-Type"]).toBe("application/json");
63
-
64
- const parsedBody = JSON.parse(calledOptions.body as string);
65
- expect(parsedBody).toEqual({
66
- query: "hello",
67
- startDate: 1000,
68
- endDate: 2000,
69
- format: "digest",
70
- });
71
- expect(result).toEqual(responseData);
72
- });
73
-
74
- it("sends format json when specified", async () => {
75
- const { searchTraces } = await import("../langwatch-api.js");
76
- mockJsonResponse({ traces: [] });
77
-
78
- await searchTraces({
79
- startDate: 1000,
80
- endDate: 2000,
81
- format: "json",
82
- });
83
-
84
- const parsedBody = JSON.parse(mockFetch.mock.calls[0]![1]!.body as string);
85
- expect(parsedBody.format).toBe("json");
86
- });
87
-
88
- describe("when response is not OK", () => {
89
- it("throws a descriptive error with status code and body", async () => {
90
- const { searchTraces } = await import("../langwatch-api.js");
91
- mockErrorResponse(401, "Unauthorized");
92
-
93
- await expect(
94
- searchTraces({ startDate: 1000, endDate: 2000 })
95
- ).rejects.toThrow("401");
96
- });
97
- });
98
- });
99
-
100
- describe("getTraceById()", () => {
101
- it("sends GET to /api/traces/{id}?format=digest by default", async () => {
102
- const { getTraceById } = await import("../langwatch-api.js");
103
- const responseData = { trace: { id: "abc" } };
104
- mockJsonResponse(responseData);
105
-
106
- const result = await getTraceById("abc");
107
-
108
- expect(mockFetch).toHaveBeenCalledWith(
109
- `${TEST_ENDPOINT}/api/traces/abc?format=digest`,
110
- expect.objectContaining({
111
- method: "GET",
112
- headers: expect.objectContaining({
113
- "X-Auth-Token": TEST_API_KEY,
114
- }),
115
- })
116
- );
117
- expect(result).toEqual(responseData);
118
- });
119
-
120
- it("sends format=json when specified", async () => {
121
- const { getTraceById } = await import("../langwatch-api.js");
122
- mockJsonResponse({});
123
-
124
- await getTraceById("abc", "json");
125
-
126
- expect(mockFetch).toHaveBeenCalledWith(
127
- `${TEST_ENDPOINT}/api/traces/abc?format=json`,
128
- expect.anything()
129
- );
130
- });
131
-
132
- it("does not include Content-Type for GET requests", async () => {
133
- const { getTraceById } = await import("../langwatch-api.js");
134
- mockJsonResponse({});
135
-
136
- await getTraceById("abc");
137
-
138
- const callHeaders = mockFetch.mock.calls[0]![1]!.headers as Record<
139
- string,
140
- string
141
- >;
142
- expect(callHeaders["Content-Type"]).toBeUndefined();
143
- });
144
- });
145
-
146
- describe("getAnalyticsTimeseries()", () => {
147
- it("sends POST to /api/analytics/timeseries", async () => {
148
- const { getAnalyticsTimeseries } = await import("../langwatch-api.js");
149
- const params = {
150
- series: [{ metric: "performance.completion_time", aggregation: "avg" }],
151
- startDate: 1000,
152
- endDate: 2000,
153
- };
154
- const responseData = { data: [] };
155
- mockJsonResponse(responseData);
156
-
157
- const result = await getAnalyticsTimeseries(params);
158
-
159
- expect(mockFetch).toHaveBeenCalledWith(
160
- `${TEST_ENDPOINT}/api/analytics/timeseries`,
161
- expect.objectContaining({
162
- method: "POST",
163
- headers: expect.objectContaining({
164
- "X-Auth-Token": TEST_API_KEY,
165
- "Content-Type": "application/json",
166
- }),
167
- body: JSON.stringify(params),
168
- })
169
- );
170
- expect(result).toEqual(responseData);
171
- });
172
- });
173
-
174
- describe("listPrompts()", () => {
175
- it("sends GET to /api/prompts", async () => {
176
- const { listPrompts } = await import("../langwatch-api.js");
177
- const responseData = [{ id: "1", name: "test" }];
178
- mockJsonResponse(responseData);
179
-
180
- const result = await listPrompts();
181
-
182
- expect(mockFetch).toHaveBeenCalledWith(
183
- `${TEST_ENDPOINT}/api/prompts`,
184
- expect.objectContaining({
185
- method: "GET",
186
- headers: expect.objectContaining({
187
- "X-Auth-Token": TEST_API_KEY,
188
- }),
189
- })
190
- );
191
- expect(result).toEqual(responseData);
192
- });
193
- });
194
-
195
- describe("getPrompt()", () => {
196
- it("sends GET to /api/prompts/{id} with encoded ID", async () => {
197
- const { getPrompt } = await import("../langwatch-api.js");
198
- const responseData = { id: "1", name: "test" };
199
- mockJsonResponse(responseData);
200
-
201
- const result = await getPrompt("my prompt/v1");
202
-
203
- expect(mockFetch).toHaveBeenCalledWith(
204
- `${TEST_ENDPOINT}/api/prompts/${encodeURIComponent("my prompt/v1")}`,
205
- expect.objectContaining({
206
- method: "GET",
207
- headers: expect.objectContaining({
208
- "X-Auth-Token": TEST_API_KEY,
209
- }),
210
- })
211
- );
212
- expect(result).toEqual(responseData);
213
- });
214
- });
215
-
216
- describe("createPrompt()", () => {
217
- it("sends POST to /api/prompts with body", async () => {
218
- const { createPrompt } = await import("../langwatch-api.js");
219
- const data = {
220
- handle: "test-prompt",
221
- messages: [{ role: "system", content: "You are helpful." }],
222
- model: "openai/gpt-4o",
223
- };
224
- const responseData = { id: "new-id", ...data };
225
- mockJsonResponse(responseData);
226
-
227
- const result = await createPrompt(data);
228
-
229
- expect(mockFetch).toHaveBeenCalledWith(
230
- `${TEST_ENDPOINT}/api/prompts`,
231
- expect.objectContaining({
232
- method: "POST",
233
- headers: expect.objectContaining({
234
- "X-Auth-Token": TEST_API_KEY,
235
- "Content-Type": "application/json",
236
- }),
237
- body: JSON.stringify(data),
238
- })
239
- );
240
- expect(result).toEqual(responseData);
241
- });
242
- });
243
-
244
- describe("updatePrompt()", () => {
245
- it("sends PUT to /api/prompts/{id} with body", async () => {
246
- const { updatePrompt } = await import("../langwatch-api.js");
247
- const data = {
248
- messages: [{ role: "system", content: "Updated" }],
249
- commitMessage: "update system prompt",
250
- };
251
- const responseData = { id: "p1", ...data };
252
- mockJsonResponse(responseData);
253
-
254
- const result = await updatePrompt("p1", data);
255
-
256
- expect(mockFetch).toHaveBeenCalledWith(
257
- `${TEST_ENDPOINT}/api/prompts/${encodeURIComponent("p1")}`,
258
- expect.objectContaining({
259
- method: "PUT",
260
- headers: expect.objectContaining({
261
- "X-Auth-Token": TEST_API_KEY,
262
- "Content-Type": "application/json",
263
- }),
264
- body: JSON.stringify(data),
265
- })
266
- );
267
- expect(result).toEqual(responseData);
268
- });
269
- });
270
-
271
- describe("getPrompt() with tag options", () => {
272
- describe("when called with tag option", () => {
273
- it("appends tag query parameter", async () => {
274
- const { getPrompt } = await import("../langwatch-api.js");
275
- mockJsonResponse({ id: "1" });
276
-
277
- await getPrompt("pizza-prompt", { tag: "production" });
278
-
279
- expect(mockFetch).toHaveBeenCalledWith(
280
- `${TEST_ENDPOINT}/api/prompts/${encodeURIComponent("pizza-prompt")}?tag=production`,
281
- expect.objectContaining({ method: "GET" })
282
- );
283
- });
284
- });
285
-
286
- describe("when called with version option", () => {
287
- it("appends version query parameter", async () => {
288
- const { getPrompt } = await import("../langwatch-api.js");
289
- mockJsonResponse({ id: "1" });
290
-
291
- await getPrompt("pizza-prompt", { version: 2 });
292
-
293
- expect(mockFetch).toHaveBeenCalledWith(
294
- `${TEST_ENDPOINT}/api/prompts/${encodeURIComponent("pizza-prompt")}?version=2`,
295
- expect.objectContaining({ method: "GET" })
296
- );
297
- });
298
- });
299
-
300
- describe("when called with no options", () => {
301
- it("sends no query string", async () => {
302
- const { getPrompt } = await import("../langwatch-api.js");
303
- mockJsonResponse({ id: "1" });
304
-
305
- await getPrompt("pizza-prompt");
306
-
307
- expect(mockFetch).toHaveBeenCalledWith(
308
- `${TEST_ENDPOINT}/api/prompts/${encodeURIComponent("pizza-prompt")}`,
309
- expect.objectContaining({ method: "GET" })
310
- );
311
- });
312
- });
313
- });
314
-
315
- describe("createPrompt() with tags", () => {
316
- describe("when called with tags", () => {
317
- it("includes tags in the request body", async () => {
318
- const { createPrompt } = await import("../langwatch-api.js");
319
- const data = {
320
- handle: "test",
321
- messages: [{ role: "system", content: "hi" }],
322
- model: "openai/gpt-5-mini",
323
- tags: ["production", "staging"],
324
- };
325
- mockJsonResponse({ id: "new" });
326
-
327
- await createPrompt(data);
328
-
329
- const body = JSON.parse(mockFetch.mock.calls[0]![1]!.body as string);
330
- expect(body.tags).toEqual(["production", "staging"]);
331
- });
332
- });
333
- });
334
-
335
- describe("updatePrompt() with tags", () => {
336
- describe("when called with tags", () => {
337
- it("includes tags in the request body", async () => {
338
- const { updatePrompt } = await import("../langwatch-api.js");
339
- mockJsonResponse({ id: "p1" });
340
-
341
- await updatePrompt("p1", {
342
- commitMessage: "tag it",
343
- tags: ["production"],
344
- });
345
-
346
- const body = JSON.parse(mockFetch.mock.calls[0]![1]!.body as string);
347
- expect(body.tags).toEqual(["production"]);
348
- });
349
- });
350
- });
351
-
352
- describe("assignPromptTag()", () => {
353
- it("sends PUT to /api/prompts/{id}/tags/{tag} with versionId", async () => {
354
- const { assignPromptTag } = await import("../langwatch-api.js");
355
- mockJsonResponse({ success: true });
356
-
357
- await assignPromptTag({ idOrHandle: "pizza-prompt", tag: "production", versionId: "v123" });
358
-
359
- const [url, opts] = mockFetch.mock.calls[0]!;
360
- expect(url).toBe(`${TEST_ENDPOINT}/api/prompts/pizza-prompt/tags/production`);
361
- expect(opts.method).toBe("PUT");
362
- expect(JSON.parse(opts.body as string)).toEqual({ versionId: "v123" });
363
- });
364
- });
365
-
366
- describe("listPromptTags()", () => {
367
- it("sends GET to /api/prompts/tags", async () => {
368
- const { listPromptTags } = await import("../langwatch-api.js");
369
- const tags = [{ id: "1", name: "production" }];
370
- mockJsonResponse(tags);
371
-
372
- const result = await listPromptTags();
373
-
374
- expect(mockFetch).toHaveBeenCalledWith(
375
- `${TEST_ENDPOINT}/api/prompts/tags`,
376
- expect.objectContaining({ method: "GET" })
377
- );
378
- expect(result).toEqual(tags);
379
- });
380
- });
381
-
382
- describe("createPromptTag()", () => {
383
- it("sends POST to /api/prompts/tags with name", async () => {
384
- const { createPromptTag } = await import("../langwatch-api.js");
385
- mockJsonResponse({ id: "t1", name: "canary" });
386
-
387
- await createPromptTag("canary");
388
-
389
- const [url, opts] = mockFetch.mock.calls[0]!;
390
- expect(url).toBe(`${TEST_ENDPOINT}/api/prompts/tags`);
391
- expect(opts.method).toBe("POST");
392
- expect(JSON.parse(opts.body as string)).toEqual({ name: "canary" });
393
- });
394
- });
395
-
396
- describe("renamePromptTag()", () => {
397
- it("sends PUT to /api/prompts/tags/{tag} with new name", async () => {
398
- const { renamePromptTag } = await import("../langwatch-api.js");
399
- mockJsonResponse({ id: "t1", name: "preview" });
400
-
401
- await renamePromptTag({ tag: "canary", name: "preview" });
402
-
403
- const [url, opts] = mockFetch.mock.calls[0]!;
404
- expect(url).toBe(`${TEST_ENDPOINT}/api/prompts/tags/canary`);
405
- expect(opts.method).toBe("PUT");
406
- expect(JSON.parse(opts.body as string)).toEqual({ name: "preview" });
407
- });
408
- });
409
-
410
- describe("deletePromptTag()", () => {
411
- it("sends DELETE to /api/prompts/tags/{tag}", async () => {
412
- const { deletePromptTag } = await import("../langwatch-api.js");
413
- mock204Response();
414
-
415
- const result = await deletePromptTag("canary");
416
-
417
- const [url, opts] = mockFetch.mock.calls[0]!;
418
- expect(url).toBe(`${TEST_ENDPOINT}/api/prompts/tags/canary`);
419
- expect(opts.method).toBe("DELETE");
420
- expect(result).toBeNull();
421
- });
422
- });
423
-
424
- describe("makeRequest()", () => {
425
- describe("when response is 204 No Content", () => {
426
- it("returns null without calling json()", async () => {
427
- const { listPromptTags } = await import("../langwatch-api.js");
428
- mock204Response();
429
-
430
- const result = await listPromptTags();
431
- expect(result).toBeNull();
432
- });
433
- });
434
- });
435
-
436
- describe("when the API returns an error", () => {
437
- it("includes the status code in the error message", async () => {
438
- const { searchTraces } = await import("../langwatch-api.js");
439
- mockErrorResponse(500, "Internal Server Error");
440
-
441
- await expect(
442
- searchTraces({ startDate: 1000, endDate: 2000 })
443
- ).rejects.toThrow("500");
444
- });
445
-
446
- it("includes the response body in the error message", async () => {
447
- const { searchTraces } = await import("../langwatch-api.js");
448
- mockErrorResponse(403, "Forbidden: invalid API key");
449
-
450
- await expect(
451
- searchTraces({ startDate: 1000, endDate: 2000 })
452
- ).rejects.toThrow("Forbidden: invalid API key");
453
- });
454
- });
455
-
456
- describe("when the API returns a handled-error JSON body", () => {
457
- const handledErrorBody = {
458
- error: "invalid_filter",
459
- message: "Filter 'llm.model' is not supported",
460
- tips: ["Use 'metadata.model' instead", "See the filter reference"],
461
- docsUrl: "https://docs.langwatch.ai/api/filters",
462
- fault: "customer",
463
- };
464
-
465
- it("parses code, tips, docsUrl and fault into LangWatchApiError fields", async () => {
466
- const { searchTraces, LangWatchApiError } = await import(
467
- "../langwatch-api.js"
468
- );
469
- mockErrorResponse(400, JSON.stringify(handledErrorBody));
470
-
471
- const error = await searchTraces({
472
- startDate: 1000,
473
- endDate: 2000,
474
- }).catch((e) => e);
475
-
476
- expect(error).toBeInstanceOf(LangWatchApiError);
477
- expect(error.status).toBe(400);
478
- expect(error.responseBody).toBe(JSON.stringify(handledErrorBody));
479
- expect(error.code).toBe("invalid_filter");
480
- expect(error.tips).toEqual(handledErrorBody.tips);
481
- expect(error.docsUrl).toBe(handledErrorBody.docsUrl);
482
- expect(error.fault).toBe("customer");
483
- });
484
-
485
- it("formats the message with the server message, Tips section and Docs link", async () => {
486
- const { searchTraces } = await import("../langwatch-api.js");
487
- mockErrorResponse(400, JSON.stringify(handledErrorBody));
488
-
489
- const error = await searchTraces({
490
- startDate: 1000,
491
- endDate: 2000,
492
- }).catch((e) => e);
493
-
494
- expect(error.message).toContain(
495
- "LangWatch API error 400: Filter 'llm.model' is not supported"
496
- );
497
- expect(error.message).toContain("Tips:");
498
- expect(error.message).toContain("- Use 'metadata.model' instead");
499
- expect(error.message).toContain("- See the filter reference");
500
- expect(error.message).toContain(
501
- "Docs: https://docs.langwatch.ai/api/filters"
502
- );
503
- });
504
-
505
- // A validation failure names the offending field AND the values it would
506
- // have taken, in `reasons[].meta`. Dropping that leaves the message's own
507
- // advice ("pick one of the types in this error's expected list") naming a
508
- // list the caller was never given.
509
- describe("when the failure carries the values it would have accepted", () => {
510
- const rejectedTypeBody = {
511
- error: "validation_error",
512
- message:
513
- "Unknown evaluator type. Pick one of the types in this error's expected list and retry.",
514
- reasons: [
515
- {
516
- code: "schema_failure",
517
- kind: "schema_failure",
518
- meta: {
519
- field: "config.evaluatorType",
520
- expected: ["ragas/response_relevancy", "langevals/llm_boolean"],
521
- received: "ragas/answer_relevancy",
522
- },
523
- },
524
- ],
525
- };
526
-
527
- /** @scenario A rejection over MCP carries the accepted types */
528
- it("keeps the reasons on the error", async () => {
529
- const { createEvaluator } = await import(
530
- "../langwatch-api-evaluators.js"
531
- );
532
- mockErrorResponse(422, JSON.stringify(rejectedTypeBody));
533
-
534
- const error = await createEvaluator({
535
- name: "Relevancy",
536
- config: { evaluatorType: "ragas/answer_relevancy" },
537
- }).catch((e) => e);
538
-
539
- expect(error.reasons).toEqual(rejectedTypeBody.reasons);
540
- });
541
-
542
- it("names the rejected field and the accepted values in the message", async () => {
543
- const { createEvaluator } = await import(
544
- "../langwatch-api-evaluators.js"
545
- );
546
- mockErrorResponse(422, JSON.stringify(rejectedTypeBody));
547
-
548
- const error = await createEvaluator({
549
- name: "Relevancy",
550
- config: { evaluatorType: "ragas/answer_relevancy" },
551
- }).catch((e) => e);
552
-
553
- expect(error.message).toContain("config.evaluatorType");
554
- expect(error.message).toContain("ragas/response_relevancy");
555
- expect(error.message).toContain("langevals/llm_boolean");
556
- expect(error.message).toContain("ragas/answer_relevancy");
557
- });
558
- });
559
-
560
- it("parses the tRPC shape using `code` instead of `error`", async () => {
561
- const { searchTraces } = await import("../langwatch-api.js");
562
- const body = {
563
- code: "dataset_not_found",
564
- message: "Dataset 'abc' not found",
565
- meta: { datasetId: "abc" },
566
- httpStatus: 404,
567
- fault: "customer",
568
- tips: ["Check the dataset slug"],
569
- docsUrl: "https://docs.langwatch.ai/datasets",
570
- reasons: [],
571
- };
572
- mockErrorResponse(404, JSON.stringify(body));
573
-
574
- const error = await searchTraces({
575
- startDate: 1000,
576
- endDate: 2000,
577
- }).catch((e) => e);
578
-
579
- expect(error.code).toBe("dataset_not_found");
580
- expect(error.tips).toEqual(["Check the dataset slug"]);
581
- expect(error.docsUrl).toBe("https://docs.langwatch.ai/datasets");
582
- expect(error.fault).toBe("customer");
583
- expect(error.message).toContain("Dataset 'abc' not found");
584
- });
585
-
586
- it("omits the Tips and Docs sections when absent", async () => {
587
- const { searchTraces } = await import("../langwatch-api.js");
588
- mockErrorResponse(
589
- 401,
590
- JSON.stringify({ error: "unauthorized", message: "Invalid API key" })
591
- );
592
-
593
- const error = await searchTraces({
594
- startDate: 1000,
595
- endDate: 2000,
596
- }).catch((e) => e);
597
-
598
- expect(error.code).toBe("unauthorized");
599
- expect(error.tips).toBeUndefined();
600
- expect(error.docsUrl).toBeUndefined();
601
- expect(error.fault).toBeUndefined();
602
- expect(error.message).toBe(
603
- "LangWatch API error 401: Invalid API key"
604
- );
605
- });
606
-
607
- it("ignores an invalid fault value", async () => {
608
- const { searchTraces } = await import("../langwatch-api.js");
609
- mockErrorResponse(
610
- 500,
611
- JSON.stringify({ error: "boom", message: "Boom", fault: "nobody" })
612
- );
613
-
614
- const error = await searchTraces({
615
- startDate: 1000,
616
- endDate: 2000,
617
- }).catch((e) => e);
618
-
619
- expect(error.fault).toBeUndefined();
620
- });
621
- });
622
-
623
- describe("when the error body is not a handled-error envelope", () => {
624
- it("keeps the raw text as the message and sets no extra fields", async () => {
625
- const { searchTraces, LangWatchApiError } = await import(
626
- "../langwatch-api.js"
627
- );
628
- mockErrorResponse(502, "<html>Bad Gateway</html>");
629
-
630
- const error = await searchTraces({
631
- startDate: 1000,
632
- endDate: 2000,
633
- }).catch((e) => e);
634
-
635
- expect(error).toBeInstanceOf(LangWatchApiError);
636
- expect(error.status).toBe(502);
637
- expect(error.responseBody).toBe("<html>Bad Gateway</html>");
638
- expect(error.message).toBe(
639
- "LangWatch API error 502: <html>Bad Gateway</html>"
640
- );
641
- expect(error.code).toBeUndefined();
642
- expect(error.tips).toBeUndefined();
643
- expect(error.docsUrl).toBeUndefined();
644
- expect(error.fault).toBeUndefined();
645
- });
646
-
647
- it("treats JSON without error/code/message fields as raw text", async () => {
648
- const { searchTraces } = await import("../langwatch-api.js");
649
- const body = JSON.stringify({ unexpected: true });
650
- mockErrorResponse(500, body);
651
-
652
- const error = await searchTraces({
653
- startDate: 1000,
654
- endDate: 2000,
655
- }).catch((e) => e);
656
-
657
- expect(error.code).toBeUndefined();
658
- expect(error.message).toBe(`LangWatch API error 500: ${body}`);
659
- });
660
- });
661
- });