@skybridge/test 2.0.0-alpha.deed804 → 2.0.0-beta.0459c30

package/README.md

@@ -6,7 +6,7 @@ check.
 
 Add the `evals` option to the Skybridge Vite plugin so `expect.chat` exists,
 then write scenarios with vitest's own `it` and `expect`. Pass the app itself
-to `start`: it is served in-process, dialing `app.fetchHandler` directly, and
+to `start`: it is served in-process, dialing an in-process handler built on `app.createServerInstance()`, and
 the tool names and argument shapes come from the app value, so no type
 parameter is needed.
 

package/dist/session-registry.d.ts

@@ -1,20 +1,17 @@
+import { type Server } from "@modelcontextprotocol/server";
 import type { LanguageModel } from "ai";
 import { Chat } from "./chat.js";
 /**
  * The minimum a Skybridge app has to expose to be served in-process: the
- * `$types` marker the assertions infer tool names from, and the fetch handler
- * the session is dialed through. Structural on purpose, so this package needs
- * no runtime dependency on `skybridge`.
+ * `$types` marker the assertions infer tool names from, and the per-request
+ * server builder the session's handler is built on. Structural on purpose, so
+ * this package needs no runtime dependency on `skybridge`.
  */
 export interface EvalApp {
     readonly $types: {
         readonly tools: object;
     };
-    readonly fetchHandler: {
-        fetch: (request: Request, options?: {
-            authInfo?: EvalIdentity;
-        }) => Promise<Response>;
-    };
+    createServerInstance(): Promise<Server>;
 }
 /**
  * The identity an in-process session claims, handed to the app as the
@@ -38,14 +35,13 @@ interface StartOptions {
 }
 /**
  * Opens a fresh MCP session and conversation against the app under test. The
- * app is served in-process, and the session is closed when the current test
+ * app is served in-process by a handler built on `app.createServerInstance`
+ * for this conversation alone, and both are closed when the current test
  * finishes, so scenarios never leak sessions and never write teardown, and
  * tests that run concurrently cannot close each other's sessions.
  *
  * The assertions are inferred from the app value itself, so `expect.chat` gets
- * the project's tool names and argument shapes with no type parameter. The
- * session talks straight to `app.fetchHandler`, which never closes: the
- * handler is memoized on the app and shared by every test.
+ * the project's tool names and argument shapes with no type parameter.
  *
  * `authInfo` claims an identity for the session: the app's per-tool scheme and
  * scope enforcement runs against it for real, only token verification is

package/dist/session-registry.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"session-registry.d.ts","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAExC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC;;;;;GAKG;AACH,MAAM,WAAW,OAAO;IACtB,QAAQ,CAAC,MAAM,EAAE;QAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAC5C,QAAQ,CAAC,YAAY,EAAE;QACrB,KAAK,EAAE,CACL,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE;YAAE,QAAQ,CAAC,EAAE,YAAY,CAAA;SAAE,KAClC,OAAO,CAAC,QAAQ,CAAC,CAAC;KACxB,CAAC;CACH;AAED;;;;GAIG;AACH,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,GAAG,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC;AAED,UAAU,YAAY;IACpB,uDAAuD;IACvD,KAAK,EAAE,aAAa,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAUD;;;;;;;;;;;;;;GAcG;AACH,wBAAsB,KAAK,CAAC,GAAG,SAAS,OAAO,EAC7C,OAAO,EAAE,YAAY,GAAG;IAAE,GAAG,EAAE,GAAG,CAAC;IAAC,QAAQ,CAAC,EAAE,YAAY,CAAA;CAAE,GAC5D,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAoBpB"}
+{"version":3,"file":"session-registry.d.ts","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAoB,KAAK,MAAM,EAAE,MAAM,8BAA8B,CAAC;AAC7E,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAExC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC;;;;;GAKG;AACH,MAAM,WAAW,OAAO;IACtB,QAAQ,CAAC,MAAM,EAAE;QAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAC5C,oBAAoB,IAAI,OAAO,CAAC,MAAM,CAAC,CAAC;CACzC;AAED;;;;GAIG;AACH,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,GAAG,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC;AAED,UAAU,YAAY;IACpB,uDAAuD;IACvD,KAAK,EAAE,aAAa,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAUD;;;;;;;;;;;;;GAaG;AACH,wBAAsB,KAAK,CAAC,GAAG,SAAS,OAAO,EAC7C,OAAO,EAAE,YAAY,GAAG;IAAE,GAAG,EAAE,GAAG,CAAC;IAAC,QAAQ,CAAC,EAAE,YAAY,CAAA;CAAE,GAC5D,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAwBpB"}

package/dist/session-registry.js

@@ -1,3 +1,4 @@
+import { createMcpHandler } from "@modelcontextprotocol/server";
 import { inject, onTestFinished } from "vitest";
 import { Chat } from "./chat.js";
 function sharedDefaults() {
@@ -10,14 +11,13 @@ function sharedDefaults() {
 }
 /**
  * Opens a fresh MCP session and conversation against the app under test. The
- * app is served in-process, and the session is closed when the current test
+ * app is served in-process by a handler built on `app.createServerInstance`
+ * for this conversation alone, and both are closed when the current test
  * finishes, so scenarios never leak sessions and never write teardown, and
  * tests that run concurrently cannot close each other's sessions.
  *
  * The assertions are inferred from the app value itself, so `expect.chat` gets
- * the project's tool names and argument shapes with no type parameter. The
- * session talks straight to `app.fetchHandler`, which never closes: the
- * handler is memoized on the app and shared by every test.
+ * the project's tool names and argument shapes with no type parameter.
  *
  * `authInfo` claims an identity for the session: the app's per-tool scheme and
  * scope enforcement runs against it for real, only token verification is
@@ -26,13 +26,17 @@ function sharedDefaults() {
 export async function start(options) {
     const config = sharedDefaults();
     const { app, authInfo } = options;
+    const handler = createMcpHandler(() => app.createServerInstance());
     const chat = await Chat.open({
         model: options.model,
         temperature: options.temperature ?? config?.temperature,
         systemPrompt: options.systemPrompt ?? config?.systemPrompt,
         maxSteps: options.maxSteps ?? config?.maxSteps,
-    }, (url, init) => app.fetchHandler.fetch(new Request(url, init), authInfo === undefined ? undefined : { authInfo }));
-    onTestFinished(() => chat.close());
+    }, (url, init) => handler.fetch(new Request(url, init), authInfo === undefined ? undefined : { authInfo }));
+    onTestFinished(async () => {
+        await chat.close();
+        await handler.close();
+    });
     return chat;
 }
 //# sourceMappingURL=session-registry.js.map

package/dist/session-registry.js.map

@@ -1 +1 @@
-{"version":3,"file":"session-registry.js","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,MAAM,QAAQ,CAAC;AAChD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAwCjC,SAAS,cAAc;IACrB,IAAI,CAAC;QACH,OAAO,MAAM,CAAC,gBAAgB,CAAC,CAAC;IAClC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;;GAcG;AACH,MAAM,CAAC,KAAK,UAAU,KAAK,CACzB,OAA6D;IAE7D,MAAM,MAAM,GAAG,cAAc,EAAE,CAAC;IAChC,MAAM,EAAE,GAAG,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC;IAElC,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,IAAI,CAC1B;QACE,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,WAAW,EAAE,OAAO,CAAC,WAAW,IAAI,MAAM,EAAE,WAAW;QACvD,YAAY,EAAE,OAAO,CAAC,YAAY,IAAI,MAAM,EAAE,YAAY;QAC1D,QAAQ,EAAE,OAAO,CAAC,QAAQ,IAAI,MAAM,EAAE,QAAQ;KAC/C,EACD,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CACZ,GAAG,CAAC,YAAY,CAAC,KAAK,CACpB,IAAI,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,EACtB,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAClD,CACJ,CAAC;IAEF,cAAc,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;IACnC,OAAO,IAAI,CAAC;AACd,CAAC"}
+{"version":3,"file":"session-registry.js","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAe,MAAM,8BAA8B,CAAC;AAE7E,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,MAAM,QAAQ,CAAC;AAChD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAmCjC,SAAS,cAAc;IACrB,IAAI,CAAC;QACH,OAAO,MAAM,CAAC,gBAAgB,CAAC,CAAC;IAClC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;GAaG;AACH,MAAM,CAAC,KAAK,UAAU,KAAK,CACzB,OAA6D;IAE7D,MAAM,MAAM,GAAG,cAAc,EAAE,CAAC;IAChC,MAAM,EAAE,GAAG,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC;IAClC,MAAM,OAAO,GAAG,gBAAgB,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,oBAAoB,EAAE,CAAC,CAAC;IAEnE,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,IAAI,CAC1B;QACE,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,WAAW,EAAE,OAAO,CAAC,WAAW,IAAI,MAAM,EAAE,WAAW;QACvD,YAAY,EAAE,OAAO,CAAC,YAAY,IAAI,MAAM,EAAE,YAAY;QAC1D,QAAQ,EAAE,OAAO,CAAC,QAAQ,IAAI,MAAM,EAAE,QAAQ;KAC/C,EACD,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CACZ,OAAO,CAAC,KAAK,CACX,IAAI,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,EACtB,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAClD,CACJ,CAAC;IAEF,cAAc,CAAC,KAAK,IAAI,EAAE;QACxB,MAAM,IAAI,CAAC,KAAK,EAAE,CAAC;QACnB,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC;IACxB,CAAC,CAAC,CAAC;IACH,OAAO,IAAI,CAAC;AACd,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@skybridge/test",
-  "version": "2.0.0-alpha.deed804",
+  "version": "2.0.0-beta.0459c30",
   "description": "Eval runner for Skybridge MCP apps: does a prompt reach the right tool with the right arguments",
   "license": "ISC",
   "type": "module",
@@ -19,11 +19,12 @@
   ],
   "dependencies": {
     "@modelcontextprotocol/client": "^2.0.0",
-    "@skybridge/vite-plugin": "2.0.0-alpha.deed804"
+    "@modelcontextprotocol/server": "^2.0.0"
   },
   "peerDependencies": {
+    "@skybridge/vite-plugin": "^2.0.0-beta.0",
     "ai": "^5.0.0",
-    "skybridge": "^2.0.0",
+    "skybridge": "^2.0.0-beta.0",
     "vitest": "^4.1.0"
   },
   "devDependencies": {
@@ -32,7 +33,8 @@
     "typescript": "^5.9.3",
     "vitest": "^4.1.10",
     "zod": "^4.4.3",
-    "skybridge": "2.0.0-alpha.deed804"
+    "@skybridge/vite-plugin": "2.0.0-beta.0459c30",
+    "skybridge": "2.0.0-beta.0459c30"
   },
   "engines": {
     "node": ">=22.23.1"