runbios-mcp 0.2.14-dev.246 → 0.2.14-dev.248

package/dist/http/mcp-handler.js.map

@@ -1 +1 @@
-{"version":3,"file":"mcp-handler.js","sourceRoot":"","sources":["../../src/http/mcp-handler.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,6BAA6B,EAAE,MAAM,oDAAoD,CAAC;AACnG,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAClD,OAAO,EAAE,mBAAmB,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AAI5D,OAAO,EAAE,mBAAmB,EAAE,MAAM,YAAY,CAAC;AAuBjD,SAAS,YAAY,CAAC,GAAa,EAAE,GAAiB;IACpD,GAAG;SACA,MAAM,CAAC,GAAG,CAAC;SACX,GAAG,CACF,kBAAkB,EAClB,6BAA6B,GAAG,CAAC,aAAa,mEAAmE,CAClH;SACA,IAAI,CAAC;QACJ,OAAO,EAAE,KAAK;QACd,KAAK,EAAE,EAAE,IAAI,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,gDAAgD,EAAE;QAClF,EAAE,EAAE,IAAI;KACT,CAAC,CAAC;AACP,CAAC;AAED,MAAM,UAAU,oBAAoB,CAAC,IAAoB;IACvD,MAAM,EAAE,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,GAAG,IAAI,CAAC;IAEnC,OAAO,KAAK,UAAU,aAAa,CAAC,GAAY,EAAE,GAAa;QAC7D,MAAM,MAAM,GAAG,GAAG,CAAC,MAAM,CAAC,eAAe,CAAC,IAAI,EAAE,CAAC;QACjD,MAAM,KAAK,GAAG,kBAAkB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAC9C,MAAM,QAAQ,GAAG,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9C,MAAM,MAAM,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,mBAAmB,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QAC5E,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,YAAY,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;YACvB,OAAO;QACT,CAAC;QACD,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,CAAC;QAEzB,MAAM,MAAM,GAAG,IAAI,UAAU,CAAC;YAC5B,OAAO,EAAE,UAAU,GAAG,CAAC,cAAc,IAAI,GAAG,CAAC,cAAc,EAAE;YAC7D,WAAW,EAAE,EAAE,aAAa,EAAE,UAAU,QAAQ,EAAE,EAAE;YACpD,WAAW,EAAE,KAAK,CAAC,YAAY;YAC/B,KAAK,EAAE,KAAK,CAAC,MAAM,IAAI,SAAS;YAChC,SAAS,EAAE,mBAAmB,OAAO,EAAE;SACxC,CAAC,CAAC;QAEH,sEAAsE;QACtE,wEAAwE;QACxE,uEAAuE;QACvE,qEAAqE;QACrE,iEAAiE;QACjE,MAAM,WAAW,GAAG,MAAM,kBAAkB,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,EAAE,GAAG,EAAE,CACpE,MAAM,CAAC,GAAG,CACR,sBAAsB,kBAAkB,CAAC,KAAK,CAAC,MAAM,CAAC,WAAW,CAClE,CACF,CAAC;QACF,IAAI,YAAY,GAAG,IAAI,GAAG,CAAC,CAAC,oBAAoB,EAAE,oBAAoB,CAAC,CAAC,CAAC;QACzE,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,GAAG,CAAmC,0BAA0B,CAAC,CAAC;YAChG,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAC;gBAAE,YAAY,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,iBAAiB,CAAC,CAAC;QACpG,CAAC;QAAC,MAAM,CAAC;YACP,KAAK,CAAC,IAAI,CAAC,mCAAmC,EAAE;gBAC9C,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,YAAY,EAAE,KAAK,CAAC,YAAY;gBAChC,QAAQ,EAAE,KAAK,CAAC,EAAE;aACnB,CAAC,CAAC;QACL,CAAC;QACD,YAAY,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC;QAEtC,MAAM,MAAM,GAAG,mBAAmB,CAAC,MAAM,EAAE;YACzC,UAAU,EAAE,CAAC,IAAI,EAAE,EAAE,CACnB,KAAK,CAAC,IAAI,CAAC,eAAe,EAAE;gBAC1B,IAAI,EAAE,IAAI,CAAC,IAAI;gBACf,EAAE,EAAE,IAAI,CAAC,EAAE;gBACX,EAAE,EAAE,IAAI,CAAC,EAAE;gBACX,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,YAAY,EAAE,KAAK,CAAC,YAAY;gBAChC,QAAQ,EAAE,KAAK,CAAC,EAAE;gBAClB,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC7C,CAAC;SACL,EAAE,SAAS,EAAE,WAAW,EAAE,YAAY,CAAC,CAAC;QACzC,MAAM,SAAS,GAAG,IAAI,6BAA6B,CAAC;YAClD,kBAAkB,EAAE,SAAS;YAC7B,kBAAkB,EAAE,IAAI;SACzB,CAAC,CAAC;QAEH,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE;YACnB,KAAK,SAAS,CAAC,KAAK,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;YACvC,KAAK,MAAM,CAAC,KAAK,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;QACtC,CAAC,CAAC,CAAC;QAEH,IAAI,CAAC;YACH,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;YAChC,MAAM,SAAS,CAAC,aAAa,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC;QACpD,CAAC;QAAC,MAAM,CAAC;YACP,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,CAAC;gBACrB,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC;oBACnB,OAAO,EAAE,KAAK;oBACd,KAAK,EAAE,EAAE,IAAI,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,uBAAuB,EAAE;oBACzD,EAAE,EAAE,IAAI;iBACT,CAAC,CAAC;YACL,CAAC;iBAAM,IAAI,CAAC,GAAG,CAAC,aAAa,EAAE,CAAC;gBAC9B,GAAG,CAAC,GAAG,EAAE,CAAC;YACZ,CAAC;QACH,CAAC;IACH,CAAC,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,gBAAgB,CAAC,GAAY,EAAE,GAAa;IAC1D,GAAG;SACA,MAAM,CAAC,GAAG,CAAC;SACX,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC;SACpB,IAAI,CAAC;QACJ,OAAO,EAAE,KAAK;QACd,KAAK,EAAE,EAAE,IAAI,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,8DAA8D,EAAE;QAChG,EAAE,EAAE,IAAI;KACT,CAAC,CAAC;AACP,CAAC"}
+{"version":3,"file":"mcp-handler.js","sourceRoot":"","sources":["../../src/http/mcp-handler.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,6BAA6B,EAAE,MAAM,oDAAoD,CAAC;AACnG,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAClD,OAAO,EAAE,mBAAmB,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AAI5D,OAAO,EAAE,mBAAmB,EAAE,MAAM,YAAY,CAAC;AAuBjD,SAAS,YAAY,CAAC,GAAa,EAAE,GAAiB;IACpD,GAAG;SACA,MAAM,CAAC,GAAG,CAAC;SACX,GAAG,CACF,kBAAkB,EAClB,6BAA6B,GAAG,CAAC,aAAa,mEAAmE,CAClH;SACA,IAAI,CAAC;QACJ,OAAO,EAAE,KAAK;QACd,KAAK,EAAE,EAAE,IAAI,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,gDAAgD,EAAE;QAClF,EAAE,EAAE,IAAI;KACT,CAAC,CAAC;AACP,CAAC;AAED,MAAM,UAAU,oBAAoB,CAAC,IAAoB;IACvD,MAAM,EAAE,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,GAAG,IAAI,CAAC;IAEnC,OAAO,KAAK,UAAU,aAAa,CAAC,GAAY,EAAE,GAAa;QAC7D,MAAM,MAAM,GAAG,GAAG,CAAC,MAAM,CAAC,eAAe,CAAC,IAAI,EAAE,CAAC;QACjD,MAAM,KAAK,GAAG,kBAAkB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAC9C,MAAM,QAAQ,GAAG,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9C,MAAM,MAAM,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,mBAAmB,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QAC5E,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,YAAY,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;YACvB,OAAO;QACT,CAAC;QACD,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,CAAC;QAEzB,MAAM,MAAM,GAAG,IAAI,UAAU,CAAC;YAC5B,OAAO,EAAE,UAAU,GAAG,CAAC,cAAc,IAAI,GAAG,CAAC,cAAc,EAAE;YAC7D,WAAW,EAAE,EAAE,aAAa,EAAE,UAAU,QAAQ,EAAE,EAAE;YACpD,WAAW,EAAE,KAAK,CAAC,YAAY;YAC/B,KAAK,EAAE,KAAK,CAAC,MAAM,IAAI,SAAS;YAChC,SAAS,EAAE,mBAAmB,OAAO,EAAE;SACxC,CAAC,CAAC;QAEH,sEAAsE;QACtE,wEAAwE;QACxE,uEAAuE;QACvE,qEAAqE;QACrE,iEAAiE;QACjE,MAAM,WAAW,GAAG,MAAM,kBAAkB,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,EAAE,GAAG,EAAE,CACpE,MAAM,CAAC,GAAG,CACR,sBAAsB,kBAAkB,CAAC,KAAK,CAAC,MAAM,CAAC,WAAW,CAClE,CACF,CAAC;QACF,IAAI,YAAY,GAAG,IAAI,GAAG,CAAC,CAAC,oBAAoB,EAAE,oBAAoB,CAAC,CAAC,CAAC;QACzE,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,GAAG,CAAmC,0BAA0B,CAAC,CAAC;YAChG,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAC;gBAAE,YAAY,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,iBAAiB,CAAC,CAAC;QACpG,CAAC;QAAC,MAAM,CAAC;YACP,KAAK,CAAC,IAAI,CAAC,mCAAmC,EAAE;gBAC9C,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,YAAY,EAAE,KAAK,CAAC,YAAY;gBAChC,QAAQ,EAAE,KAAK,CAAC,EAAE;aACnB,CAAC,CAAC;QACL,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,CAAC,yBAAyB,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,wBAAwB,CAAC,EAAE,CAAC;YAC1H,YAAY,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QAC5B,CAAC;QACD,YAAY,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC;QAEtC,MAAM,MAAM,GAAG,mBAAmB,CAAC,MAAM,EAAE;YACzC,UAAU,EAAE,CAAC,IAAI,EAAE,EAAE,CACnB,KAAK,CAAC,IAAI,CAAC,eAAe,EAAE;gBAC1B,IAAI,EAAE,IAAI,CAAC,IAAI;gBACf,EAAE,EAAE,IAAI,CAAC,EAAE;gBACX,EAAE,EAAE,IAAI,CAAC,EAAE;gBACX,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,YAAY,EAAE,KAAK,CAAC,YAAY;gBAChC,QAAQ,EAAE,KAAK,CAAC,EAAE;gBAClB,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC7C,CAAC;SACL,EAAE,SAAS,EAAE,WAAW,EAAE,YAAY,CAAC,CAAC;QACzC,MAAM,SAAS,GAAG,IAAI,6BAA6B,CAAC;YAClD,kBAAkB,EAAE,SAAS;YAC7B,kBAAkB,EAAE,IAAI;SACzB,CAAC,CAAC;QAEH,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE;YACnB,KAAK,SAAS,CAAC,KAAK,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;YACvC,KAAK,MAAM,CAAC,KAAK,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;QACtC,CAAC,CAAC,CAAC;QAEH,IAAI,CAAC;YACH,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;YAChC,MAAM,SAAS,CAAC,aAAa,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC;QACpD,CAAC;QAAC,MAAM,CAAC;YACP,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,CAAC;gBACrB,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC;oBACnB,OAAO,EAAE,KAAK;oBACd,KAAK,EAAE,EAAE,IAAI,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,uBAAuB,EAAE;oBACzD,EAAE,EAAE,IAAI;iBACT,CAAC,CAAC;YACL,CAAC;iBAAM,IAAI,CAAC,GAAG,CAAC,aAAa,EAAE,CAAC;gBAC9B,GAAG,CAAC,GAAG,EAAE,CAAC;YACZ,CAAC;QACH,CAAC;IACH,CAAC,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,gBAAgB,CAAC,GAAY,EAAE,GAAa;IAC1D,GAAG;SACA,MAAM,CAAC,GAAG,CAAC;SACX,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC;SACpB,IAAI,CAAC;QACJ,OAAO,EAAE,KAAK;QACd,KAAK,EAAE,EAAE,IAAI,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,8DAA8D,EAAE;QAChG,EAAE,EAAE,IAAI;KACT,CAAC,CAAC;AACP,CAAC"}

package/dist/server.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../src/server.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AAkDpE,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAClD,OAAO,EAAsB,KAAK,WAAW,EAAE,MAAM,aAAa,CAAC;AACnE,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AAEvC,OAAO,EAAE,OAAO,EAAE,CAAC;AAEnB,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,EAAE,EAAE,OAAO,CAAC;IACZ,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,WAAW;IAC1B,UAAU,CAAC,EAAE,CAAC,IAAI,EAAE,YAAY,KAAK,IAAI,CAAC;CAC3C;AAUD,MAAM,WAAW,sBAAsB;IAErC,cAAc,CAAC,EAAE,OAAO,CAAC;IAEzB,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB;AA6CD,wBAAgB,6BAA6B,CAAC,IAAI,EAAE,OAAO,GAAG,OAAO,CA0BpE;AAyWD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AACH,wBAAgB,8BAA8B,CAAC,IAAI,EAAE,OAAO,GAAG,OAAO,CAkDrE;AA0BD;;;;;;;;;GASG;AACH,wBAAgB,wBAAwB,CAAC,IAAI,EAAE,OAAO,GAAG,OAAO,CAiD/D;AAID,wBAAgB,mBAAmB,CACjC,MAAM,EAAE,UAAU,EAClB,KAAK,CAAC,EAAE,WAAW,EACnB,cAAc,CAAC,EAAE,sBAAsB,EAOvC,WAAW,GAAE,WAAkC,EAC/C,YAAY,CAAC,EAAE,WAAW,CAAC,MAAM,CAAC,GACjC,SAAS,CA27GX"}
+{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../src/server.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AAkDpE,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAClD,OAAO,EAAsB,KAAK,WAAW,EAAE,MAAM,aAAa,CAAC;AACnE,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AAEvC,OAAO,EAAE,OAAO,EAAE,CAAC;AAEnB,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,EAAE,EAAE,OAAO,CAAC;IACZ,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,WAAW;IAC1B,UAAU,CAAC,EAAE,CAAC,IAAI,EAAE,YAAY,KAAK,IAAI,CAAC;CAC3C;AAUD,MAAM,WAAW,sBAAsB;IAErC,cAAc,CAAC,EAAE,OAAO,CAAC;IAEzB,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB;AA6CD,wBAAgB,6BAA6B,CAAC,IAAI,EAAE,OAAO,GAAG,OAAO,CA0BpE;AAyWD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AACH,wBAAgB,8BAA8B,CAAC,IAAI,EAAE,OAAO,GAAG,OAAO,CAkDrE;AA0BD;;;;;;;;;GASG;AACH,wBAAgB,wBAAwB,CAAC,IAAI,EAAE,OAAO,GAAG,OAAO,CAiD/D;AAID,wBAAgB,mBAAmB,CACjC,MAAM,EAAE,UAAU,EAClB,KAAK,CAAC,EAAE,WAAW,EACnB,cAAc,CAAC,EAAE,sBAAsB,EAOvC,WAAW,GAAE,WAAkC,EAC/C,YAAY,CAAC,EAAE,WAAW,CAAC,MAAM,CAAC,GACjC,SAAS,CAqgHX"}

package/dist/server.js

@@ -625,8 +625,8 @@ launchGates = resolveLaunchGates(), allowedTools) {
             + prelaunchNote
             + "Rules that no tool schema can express:\n"
             + "1. Identity: every tool acts only in the API key's bound organization and workspace. Call introspect_api_key first; never invent or substitute an org/workspace/account.\n"
-            + "2. Serverless is the default for immediate inference: no GPU booking, no deployment object, per-token billing. Use a published model id and chat_with_inference. Do NOT call create_inference for serverless.\n"
-            + "3. Dedicated inference reserves GPUs and bills per second. Use get_inference_gpu_options -> preflight_inference -> explicit spend consent -> create_inference -> poll get_inference_status until running -> chat_with_inference.\n"
+            + "2. Serverless is the default for immediate inference: no GPU booking, no deployment object, per-token billing. Use chat_with_inference or message_with_inference with a published model that supports its dialect. Serverless has no completions, embeddings, or rerank route. Do NOT call create_inference for serverless.\n"
+            + "3. Dedicated inference reserves GPUs and bills per second. Use get_inference_gpu_options -> preflight_inference -> explicit spend consent -> create_inference -> poll get_inference_status until running -> invoke the task this deployment actually serves (chat, messages, completions, embeddings, or rerank).\n"
             + (launchGates.training
                 ? "4. Models: list_models / search_models identify hosted base models for a new dedicated deployment. "
                 : "4. Models: list_models / search_models identify hosted base models for training and dedicated deployment. ")
@@ -635,7 +635,7 @@ launchGates = resolveLaunchGates(), allowedTools) {
             + `6. Context window: the default is min(model native max, ${CONTEXT_DEFAULT_CEILING}) tokens, falling back to `
             + `${CONTEXT_UNKNOWN_FALLBACK} when unknown. Read model.native_max_context from preflight_inference.\n`
             + "7. GPUs and money: never substitute a GPU, create a paid resource, or exceed a price cap without explicit user approval. A 503 is transient, never proof of no capacity.\n"
-            + "8. Report a dedicated endpoint as live only when get_inference_status says running; then call it with chat_with_inference. Serverless needs no such poll.";
+            + "8. Report a dedicated endpoint as live only when get_inference_status says running; call only the verb its model serves, never guess an unsupported mode. Serverless needs no such poll.";
     const mcp = new McpServer({
         name: "Run BiOS",
         version: VERSION,
@@ -781,7 +781,7 @@ ${names.map((name) => `- ${name}`).join("\n")}
 
 ${gatedSurfaces.join(" and ")} ${gatedSurfaces.length > 1 ? "are" : "is"} not available in this build yet — the creation tools are not registered, so no tool call (and no workaround) can start one today. Do not retry or search for one.
 
-What still works: every read and lifecycle tool for existing datasets and training jobs (list_datasets, get_dataset_status, preview_dataset, update_dataset_column_mapping, revalidate_dataset, update_dataset_source, set_dataset_integration, delete_dataset, list_training_jobs, get_training_status, get_training_metrics, get_training_logs, get_training_evals, stop_training_job, resume_training_job, get_training_checkpoints, get_checkpoint_download_manifest, delete_training_checkpoint), and the full inference surface (preflight_inference, create_inference, get_inference_status, chat_with_inference).`;
+What still works: every read and lifecycle tool for existing datasets and training jobs (list_datasets, get_dataset_status, preview_dataset, update_dataset_column_mapping, revalidate_dataset, update_dataset_source, set_dataset_integration, delete_dataset, list_training_jobs, get_training_status, get_training_metrics, get_training_logs, get_training_evals, stop_training_job, resume_training_job, get_training_checkpoints, get_checkpoint_download_manifest, delete_training_checkpoint), and the full inference surface (preflight_inference, create_inference, get_inference_status, chat_with_inference, message_with_inference, complete_with_inference, embed_with_inference, rerank_with_inference).`;
         const gatedOverview = `# Run BiOS Platform
 
 Run BiOS has TWO inference products. Choosing the right one is step zero; they are not two names for the same operation.
@@ -790,8 +790,8 @@ Run BiOS has TWO inference products. Choosing the right one is step zero; they a
 Use this by default for published Run BiOS model ids. There is no GPU selection, booking, deployment object, provisioning wait, stop/delete lifecycle, or hourly price cap. It is routed by the platform and billed per token.
 1. Call introspect_api_key to confirm the bound workspace and serverless scope.
 2. Call list_inference_models with a serverless-scoped key, inspect usf_unreachable_sources, then get_inference_model for the id and serving capabilities. list_models is the different trainable registry; do not substitute its repo id for a serverless slug.
-3. Call chat_with_inference with that model id, messages, and only capabilities the model advertises. Streaming is optional.
-4. The completion is the result. Do NOT call preflight_inference or create_inference for serverless.
+3. Call chat_with_inference or message_with_inference with that model id and only capabilities the model advertises. Streaming is optional; message_with_inference returns ordered Anthropic events when streaming.
+4. Serverless offers chat and Anthropic Messages, not /v1/completions, /v1/embeddings, or /v1/rerank. Do NOT call preflight_inference or create_inference for serverless.
 5. If this credential has analytics:read, use get_serverless_limits and get_serverless_usage_daily for workspace RPM/spend settings and full-history admitted usage; get_serverless_usage_overview, get_serverless_usage_by_model, list_serverless_requests and get_serverless_usage_timeseries show retained request detail. These never reveal another workspace or per-key usage.
 
 ## Dedicated deployment — reserved GPU, durable endpoint
@@ -803,7 +803,7 @@ Choose this for dedicated capacity, custom/trained checkpoints, or a durable end
 5. **Ask for approval** — before create_inference, state the chosen GPU, count, hourly cap, storage and queue consent. Never substitute an alternative silently.
 6. **Create** — create_inference; retain the returned deployment id. MCP deliberately withholds the one-time dedicated key because tool text is model-visible; use the console or direct SDK/API when a standalone key is required. A booking handle must be polled with get_inference_booking.
 7. **Wait for reality** — poll get_inference_status through queue/provisioning/loading; never report the endpoint live until status=running.
-8. **Run** — call chat_with_inference using the workspace key and deployment name/id.
+8. **Run** — call chat_with_inference using the workspace key and deployment name/id when this deployment serves chat. For another declared serving task choose message_with_inference, complete_with_inference, embed_with_inference, or rerank_with_inference; never infer a mode from the model name.
 9. **Observe/control/clean up** — metrics, notifications, stop/resume/restart, then delete_inference when the user explicitly asks to remove the paid deployment.
 
 ## Launching soon — not offered in this build
@@ -866,8 +866,9 @@ get_inference_status instead of asserting them.
 8. get_inference_notifications - Inspect durable email delivery, retries, and dead letters
 9. update_inference_policy - Change explicit queue consent or the maximum accepted hourly price
 10. stop_inference / resume_inference / restart_inference - Control serving lifecycle
-11. chat_with_inference - Call a serverless catalog model by id or a dedicated deployment via the unified /v1 endpoint (optional streaming)
-12. delete_inference - Permanently remove a deployment after verified infrastructure teardown
+11. chat_with_inference / message_with_inference - Call a compatible serverless catalog model or dedicated deployment; chat aggregates SSE, Messages returns ordered task events
+12. complete_with_inference / embed_with_inference / rerank_with_inference - Dedicated deployments only, and only when the model advertises that task; completions can return bounded ordered SSE events
+13. delete_inference - Permanently remove a deployment after verified infrastructure teardown
 
 ## Serverless Workspace Observability (analytics:read)
 - get_serverless_limits - Read saved workspace RPM and monthly spend cap; the org's tier may clamp effective admission further
@@ -1003,8 +1004,9 @@ get_inference_status instead of asserting them.
 8. get_inference_notifications - Inspect durable email delivery, retries, and dead letters
 9. update_inference_policy - Change explicit queue consent or the maximum accepted hourly price
 10. stop_inference / resume_inference / restart_inference - Control serving lifecycle
-11. chat_with_inference - Call a serverless catalog model by id or a dedicated deployment via the unified /v1 endpoint (optional streaming)
-12. delete_inference - Permanently remove a deployment after verified infrastructure teardown
+11. chat_with_inference / message_with_inference - Call a compatible serverless catalog model or dedicated deployment; chat aggregates SSE, Messages returns ordered task events
+12. complete_with_inference / embed_with_inference / rerank_with_inference - Dedicated deployments only, and only when the model advertises that task; completions can return bounded ordered SSE events
+13. delete_inference - Permanently remove a deployment after verified infrastructure teardown
 
 ## Serverless Workspace Observability (analytics:read)
 - get_serverless_limits - Read saved workspace RPM and monthly spend cap; the org's tier may clamp effective admission further
@@ -1706,7 +1708,7 @@ create (1-5 choices), and the queue itself stays opt-in.`,
     /* ══════════════════════════════════════════════════════════════════════════ */
     /*  TOOL: get_recommended_gpu                                               */
     /* ══════════════════════════════════════════════════════════════════════════ */
-    server.tool("get_recommended_gpu", "Get authoritative, model-aware GPU options from the training control plane. Returns required counts, live availability, total hourly prices, the cheapest currently bookable recommendation, and compatible alternatives when no GPU is available. An option's host_ram_gb is the conservative per-GPU host-RAM lower bound; min_host_ram_gb is the total minimum requested from the provider. Generic GPU stock does not prove high-RAM stock: when bookable=false for a compatible option, queue that exact shape unbilled rather than recommending another unverified GPU. Pass effective_batch (samples per optimizer step -- the quality decision, 64 is standard) and max_length, and every option also carries the per-device split that runs it there: recommended_micro_batch x recommended_grad_accum x required_count. Use that split as per_device_train_batch_size / gradient_accumulation_steps when creating the job. It does not invent a duration or total-cost estimate.", {
+    server.tool("get_recommended_gpu", "Get authoritative, model-aware GPU options from the training control plane. Returns required counts, live availability, total hourly prices, the cheapest currently bookable recommendation, and compatible alternatives when no GPU is available. An option's host_ram_gb is the conservative per-GPU host-RAM lower bound; min_host_ram_gb is the total minimum requested from the provider. Generic GPU stock does not prove high-RAM stock: when bookable=false for a compatible option, queue that exact shape unbilled rather than recommending another unverified GPU. Pass effective_batch (samples per optimizer step -- the quality decision, 64 is standard) and max_length, and every option also carries the per-device split that runs it there: recommended_micro_batch x recommended_grad_accum x required_count. Use that split as per_device_train_batch_size / gradient_accumulation_steps when creating the job. It does not invent a duration or total-cost estimate. If method_unavailable or adapter_unavailable is true, this method and adapter have no verified offer: do not queue or create. The model_support.admissible_method_adapters map lists the method-specific adapter evidence; a declared capability alone is not proof of a completed run.", {
         model: z.string().describe("Full model name (e.g., 'meta-llama/Llama-3.1-8B-Instruct')"),
         model_revision: z.string().optional().describe("Pinned model commit, branch or tag; the answer returns the exact immutable revision."),
         integration_id: z.string().optional().describe("Connected Hugging Face integration for gated model sizing."),
@@ -2339,7 +2341,7 @@ create (1-5 choices), and the queue itself stays opt-in.`,
     /* ══════════════════════════════════════════════════════════════════════════ */
     /*  TOOL: get_training_capabilities                                          */
     /* ══════════════════════════════════════════════════════════════════════════ */
-    server.tool("get_training_capabilities", "Return the authoritative training contract Run BiOS enforces today: enabled and disabled methods, algorithms and adapters; exact hyperparameter names, aliases, types, defaults, enums and ranges; dependencies and incompatibilities. contract_version/schema_version identify the contract you are reading. Call this before constructing an automated training payload instead of relying on a hard-coded client list.", {}, async () => {
+    server.tool("get_training_capabilities", "Return the authoritative training contract Run BiOS enforces today: enabled and disabled methods, algorithms and adapters; exact hyperparameter names, aliases, types, defaults, enums and ranges; dependencies and incompatibilities. contract_version/schema_version identify the contract you are reading. Call this before constructing an automated training payload instead of relying on a hard-coded client list. Engine support is not per-model evidence: inspect get_recommended_gpu.model_support.admissible_method_adapters, then preflight_training_job, before offering or creating any method and adapter combination.", {}, async () => {
         const data = await client.api("/api/training/capabilities");
         // The contract used to carry image_compatibility — the training-engine image
         // VERSION range it was written against. training-service no longer emits it;
@@ -2688,6 +2690,53 @@ create (1-5 choices), and the queue itself stays opt-in.`,
             return json(data);
         });
     }
+    const taskControlSchema = {
+        model: z.string().min(1).optional().describe("Serverless catalog model id, or dedicated deployment name/id. A dedicated inference key may select its deployment without a model id."),
+        idempotency_key: z.string().min(1).max(255).optional().describe("Caller-generated key for an explicit retry of the same POST; never retried automatically."),
+        request_id: z.string().min(1).max(255).optional(),
+    };
+    const taskStreamSchema = z.boolean().optional().describe("Consume SSE and return the ordered, unmodified events in one MCP result (maximum 2 MB or 4096 events). Defaults to false; use an SDK for longer streams.");
+    server.tool("complete_with_inference", "Call a dedicated deployment that serves OpenAI text completions on /v1/completions. Serverless pools do not expose this route. Use a model/deployment that advertises this task. stream=true returns the ordered SSE events. Never implicitly retries a billable POST.", {
+        ...taskControlSchema,
+        prompt: z.union([z.string().min(1), z.array(z.string().min(1)).min(1)]),
+        max_tokens: z.number().int().min(1).max(CONTEXT_TOOL_MAX).optional().describe(MAX_TOKENS_POLICY_TEXT),
+        temperature: z.number().min(0).max(2).optional(),
+        top_p: z.number().min(0).max(1).optional(),
+        stop: z.union([z.string(), z.array(z.string())]).optional(),
+        stream: taskStreamSchema,
+    }, async (input, extra) => {
+        const { idempotency_key, request_id, ...body } = input;
+        return json(await client.inferenceApi(body, idempotency_key, request_id, extra.signal, "/v1/completions"));
+    });
+    server.tool("embed_with_inference", "Create embeddings with a dedicated deployment that serves /v1/embeddings. The serverless pool has no embeddings route; a chat-only model cannot embed. Never implicitly retries a billable POST.", {
+        ...taskControlSchema,
+        input: z.union([z.string().min(1), z.array(z.string().min(1)).min(1)]),
+    }, async (input, extra) => {
+        const { idempotency_key, request_id, ...body } = input;
+        return json(await client.inferenceApi(body, idempotency_key, request_id, extra.signal, "/v1/embeddings"));
+    });
+    server.tool("rerank_with_inference", "Rerank documents with a dedicated deployment that serves /v1/rerank. The serverless pool has no rerank route; a chat-only model cannot rerank. Never implicitly retries a billable POST.", {
+        ...taskControlSchema,
+        query: z.string().min(1),
+        documents: z.array(z.string().min(1)).min(1),
+    }, async (input, extra) => {
+        const { idempotency_key, request_id, ...body } = input;
+        return json(await client.inferenceApi(body, idempotency_key, request_id, extra.signal, "/v1/rerank"));
+    });
+    server.tool("message_with_inference", "Call an Anthropic-compatible /v1/messages model: a capable serverless catalog model or a dedicated deployment that serves this dialect. The selected inference credential stays server-side. stream=true returns ordered Anthropic SSE events without losing tool-use deltas; never implicitly retries a billable POST.", {
+        ...taskControlSchema,
+        messages: z.array(z.record(z.string(), z.unknown())).min(1),
+        max_tokens: z.number().int().min(1).max(CONTEXT_TOOL_MAX).describe(MAX_TOKENS_POLICY_TEXT),
+        system: z.union([z.string(), z.array(z.record(z.string(), z.unknown()))]).optional(),
+        tools: z.array(z.record(z.string(), z.unknown())).optional(),
+        tool_choice: z.record(z.string(), z.unknown()).optional(),
+        stop_sequences: z.array(z.string()).optional(),
+        thinking: z.record(z.string(), z.unknown()).optional(),
+        stream: taskStreamSchema,
+    }, async (input, extra) => {
+        const { idempotency_key, request_id, ...body } = input;
+        return json(await client.inferenceApi(body, idempotency_key, request_id, extra.signal, "/v1/messages"));
+    });
     const serverlessUsageWindow = z.enum(["1h", "24h", "7d", "30d", "90d"]);
     const serverlessUsageWindowInput = serverlessUsageWindow.optional().describe("Analytics window (default 24h; for full-history admitted totals use daily). Raw request details are retained for about seven days.");
     server.tool("get_serverless_limits", "Read the current workspace's saved RPM ceiling and monthly spend cap in cents. analytics:read and current workspace access are required. This does not change a limit or return key details; the org's tier may clamp the effective RPM further.", {}, async () => json(await client.serverlessRead("/api/serverless/limits", "limits")));