sunpeak 0.20.2 → 0.20.6

package/bin/lib/eval/eval-runner.mjs

@@ -112,9 +112,35 @@ export async function discoverAndConvertTools(client) {
   const tools = {};
 
   for (const t of mcpTools) {
+    // Clean up the MCP inputSchema for AI provider compatibility.
+    // OpenAI rejects $schema, additionalProperties: {} (empty schema has no type),
+    // and other JSON Schema features that MCP servers may include.
+    const rawSchema = t.inputSchema || { type: 'object', properties: {} };
+    const cleanSchema = { ...rawSchema };
+    delete cleanSchema.$schema;
+    if (
+      cleanSchema.additionalProperties != null &&
+      typeof cleanSchema.additionalProperties === 'object' &&
+      Object.keys(cleanSchema.additionalProperties).length === 0
+    ) {
+      // Empty additionalProperties ({}) causes OpenAI to report type: "None".
+      // Remove it so the schema is a plain { type: "object", properties: {...} }.
+      delete cleanSchema.additionalProperties;
+    }
+    if (!cleanSchema.type) cleanSchema.type = 'object';
+    if (!cleanSchema.properties) cleanSchema.properties = {};
+    // Remove `required` so the model isn't forced to ask the user for every
+    // parameter before calling a tool. Eval prompts are intentionally vague
+    // ("show me photo albums") and the model should call the tool with
+    // reasonable defaults, not refuse because required fields are missing.
+    delete cleanSchema.required;
+
     tools[t.name] = aiTool({
       description: t.description || '',
-      parameters: jsonSchema(t.inputSchema || { type: 'object', properties: {} }),
+      // Set both so the tool works with ai v4/v5 (reads `parameters`)
+      // and ai v6 (reads `inputSchema`). tool() passes through both.
+      inputSchema: jsonSchema(cleanSchema),
+      parameters: jsonSchema(cleanSchema),
       execute: async (args) => {
         const result = await client.callTool({ name: t.name, arguments: args });
         // Return a simplified version for the model to consume

package/bin/lib/eval/model-registry.mjs

@@ -44,7 +44,12 @@ export async function resolveModel(modelId) {
   // that creates model instances: openai('gpt-4o'), anthropic('claude-...'), google('gemini-...')
   if (pkg === '@ai-sdk/openai') {
     const { openai } = provider;
-    return openai(modelId);
+    // @ai-sdk/openai v3 defaults to the Responses API, which requires strict
+    // JSON Schema (additionalProperties: false at every level, all properties
+    // required) — incompatible with arbitrary MCP server schemas. Use .chat()
+    // (Chat Completions API) when available. v1/v2 default to Chat Completions
+    // already and may not have .chat(), so fall back to the default.
+    return typeof openai.chat === 'function' ? openai.chat(modelId) : openai(modelId);
   }
   if (pkg === '@ai-sdk/anthropic') {
     const { anthropic } = provider;

package/bin/lib/test/base-config.mjs

@@ -5,8 +5,6 @@
  * Produces a config with per-host Playwright projects, sensible defaults for
  * MCP App testing, and a webServer entry to launch the inspector backend.
  */
-import { getPortSync } from '../get-port.mjs';
-
 /**
  * @param {Object} options
  * @param {string[]} options.hosts - Host shells to create projects for
@@ -63,10 +61,19 @@ export function createBaseConfig({ hosts, testDir, webServer, port, use, globalS
 /**
  * Resolve ports for the inspector and sandbox proxy.
  * Respects env vars for CI where validate.mjs assigns unique ports.
+ *
+ * Uses FIXED default ports (no dynamic probing) so all Playwright workers
+ * resolve the same baseURL. Dynamic port probing (getPortSync) caused flaky
+ * tests: the main process would pick port X, start the webServer on it, then
+ * worker processes re-evaluating the config would find X occupied and resolve
+ * to random ports Y/Z — causing ERR_CONNECTION_REFUSED.
+ *
+ * If the default port is busy, Playwright's reuseExistingServer (local) reuses
+ * it, or strictPort (CI) fails fast with a clear error.
  */
 export function resolvePorts() {
-  const port = parsePort(process.env.SUNPEAK_TEST_PORT) ?? getPortSync(6776);
-  const sandboxPort = parsePort(process.env.SUNPEAK_SANDBOX_PORT) ?? getPortSync(24680);
+  const port = parsePort(process.env.SUNPEAK_TEST_PORT) ?? 6776;
+  const sandboxPort = parsePort(process.env.SUNPEAK_SANDBOX_PORT) ?? 24680;
   return { port, sandboxPort };
 }
 

package/bin/sunpeak.js

@@ -102,22 +102,11 @@ function getVersion() {
       {
         const resources = discoverResources();
         console.log(`
-☀️ 🏔️ sunpeak - Inspector, testing framework, and app framework for MCP Apps
+☀️ 🏔️ sunpeak - App framework, testing framework, and inspector for MCP Apps
 
-Install:
-  pnpm add -g sunpeak
+Usage: npx sunpeak <command>
 
-Testing (works with any MCP server):
-  sunpeak inspect          Inspect any MCP server in the inspector
-    --server, -s <url|cmd> MCP server URL or stdio command (required)
-    --simulations <dir>    Simulation JSON directory
-  sunpeak test             Run e2e tests against the inspector
-    init                   Scaffold test infrastructure into a project
-    --unit                 Run unit tests (vitest)
-    --live                 Run live tests against real hosts
-    --eval                 Run evals against LLM models
-
-App framework (for sunpeak projects):
+App framework:
   sunpeak new [name] [resources]  Create a new project
   sunpeak dev              Start dev server + inspector + MCP endpoint
     --no-begging           Suppress GitHub star message
@@ -125,8 +114,20 @@ App framework (for sunpeak projects):
   sunpeak start            Start production MCP server
     --port, -p             Server port (default: 8000, or PORT env)
   sunpeak upgrade          Upgrade sunpeak to latest version
-  sunpeak --version        Show version number
 
+Testing (works with any MCP server):
+  sunpeak test             Run e2e tests against the inspector
+    init                   Scaffold test infrastructure into a project
+    --unit                 Run unit tests (vitest)
+    --live                 Run live tests against real hosts
+    --eval                 Run evals against LLM models
+
+Inspector (works with any MCP server):
+  sunpeak inspect          Inspect any MCP server in the inspector
+    --server, -s <url|cmd> MCP server URL or stdio command (required)
+    --simulations <dir>    Simulation JSON directory
+
+  sunpeak --version        Show version number
   Resources: ${resources.join(', ')} (comma/space separated)
   Example: sunpeak new sunpeak-app "${resources.slice(0, 2).join(',')}"
 `);

package/dist/chatgpt/index.cjs

@@ -1,6 +1,6 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 const require_chunk = require("../chunk-9hOWP6kD.cjs");
-const require_inspector = require("../inspector-DAA1Wiyh.cjs");
+const require_inspector = require("../inspector-D0qOqYX2.cjs");
 const require_inspector_url = require("../inspector-url-C3LTKgXt.cjs");
 const require_discovery = require("../discovery-Clu4uHp1.cjs");
 //#region src/chatgpt/index.ts

package/dist/chatgpt/index.js

@@ -1,5 +1,5 @@
 import { r as __exportAll } from "../chunk-D6g4UhsZ.js";
-import { _ as McpAppHost, d as ThemeProvider, f as useThemeContext, g as extractResourceCSP, h as IframeResource, n as resolveServerToolResult, t as Inspector, v as SCREEN_WIDTHS } from "../inspector-BBDa5yCm.js";
+import { _ as McpAppHost, d as ThemeProvider, f as useThemeContext, g as extractResourceCSP, h as IframeResource, n as resolveServerToolResult, t as Inspector, v as SCREEN_WIDTHS } from "../inspector-60Na_Zc4.js";
 import { t as createInspectorUrl } from "../inspector-url-CyQcuBI9.js";
 import { c as toPascalCase, i as findResourceKey, n as extractSimulationKey, r as findResourceDirs, s as getComponentName, t as extractResourceKey } from "../discovery-Cgoegt62.js";
 //#region src/chatgpt/index.ts

package/dist/claude/index.cjs

@@ -1,4 +1,4 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 require("../chunk-9hOWP6kD.cjs");
-const require_inspector = require("../inspector-DAA1Wiyh.cjs");
+const require_inspector = require("../inspector-D0qOqYX2.cjs");
 exports.Inspector = require_inspector.Inspector;

package/dist/claude/index.js

@@ -1,2 +1,2 @@
-import { t as Inspector } from "../inspector-BBDa5yCm.js";
+import { t as Inspector } from "../inspector-60Na_Zc4.js";
 export { Inspector };

package/dist/host/chatgpt/index.cjs

@@ -1,6 +1,6 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 require("../../chunk-9hOWP6kD.cjs");
-const require_use_app = require("../../use-app-DPkj5Jp_.cjs");
+const require_use_app = require("../../use-app-B33mckz4.cjs");
 let react = require("react");
 //#region src/host/chatgpt/openai-types.ts
 /**

package/dist/host/chatgpt/index.js

@@ -1,4 +1,4 @@
-import { t as useApp } from "../../use-app-Cr0auUa1.js";
+import { t as useApp } from "../../use-app-kv5GQr0G.js";
 import { useCallback } from "react";
 //#region src/host/chatgpt/openai-types.ts
 /**

package/dist/index.cjs

@@ -1,15 +1,15 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 const require_chunk = require("./chunk-9hOWP6kD.cjs");
 const require_protocol = require("./protocol-C8pFDmcy.cjs");
-const require_use_app = require("./use-app-DPkj5Jp_.cjs");
-const require_inspector = require("./inspector-DAA1Wiyh.cjs");
+const require_use_app = require("./use-app-B33mckz4.cjs");
+const require_inspector = require("./inspector-D0qOqYX2.cjs");
 const require_host_index = require("./host/index.cjs");
 const require_inspector_index = require("./inspector/index.cjs");
 const require_chatgpt_index = require("./chatgpt/index.cjs");
 let react = require("react");
 react = require_chunk.__toESM(react, 1);
 let react_jsx_runtime = require("react/jsx-runtime");
-//#region ../../node_modules/.pnpm/@modelcontextprotocol+ext-apps@1.5.0_@modelcontextprotocol+sdk@1.29.0_zod@4.3.6__react-_a8a8e071c354e7dd6f62871eadf46f99/node_modules/@modelcontextprotocol/ext-apps/dist/src/react/index.js
+//#region ../../node_modules/.pnpm/@modelcontextprotocol+ext-apps@1.6.0_@modelcontextprotocol+sdk@1.29.0_zod@4.3.6__react-_1fd7d0151a0915598274278e5ddb69e9/node_modules/@modelcontextprotocol/ext-apps/dist/src/react/index.js
 var m = require_protocol.union([require_protocol.literal("light"), require_protocol.literal("dark")]).describe("Color theme preference for the host environment."), N$1 = require_protocol.union([
 	require_protocol.literal("inline"),
 	require_protocol.literal("fullscreen"),