chore: use playwright-mcp flag to omit images

Author: Burry

src/_internal/browser-eval-manager.ts

@@ -63,7 +63,7 @@ export async function startBrowserEvalMCP(options?: {
   console.error("[Browser Eval Manager] Starting playwright-mcp server with verbose logging...")
 
   // Build args for playwright-mcp
-  const args: string[] = ["@playwright/mcp@latest"]
+  const args: string[] = ["@playwright/mcp@latest", "--image-responses", "omit"]
 
   if (options?.browser) {
     args.push("--browser", options.browser)

src/tools/browser-eval.ts

@@ -149,45 +149,6 @@ type BrowserEvalArgs = {
   files?: string[]
 }
 
-/**
- * Filter out base64 image data from the `screenshot` result to reduce context usage.
- * Extract file paths from text content and attach guidance.
- */
-export function filterImageDataFromResult(result: unknown): unknown {
-  if (typeof result !== "object" || result === null) {
-    return result
-  }
-
-  const typedResult = result as { content?: { type: string; text?: string; data?: string }[] }
-
-  if (!Array.isArray(typedResult.content)) {
-    return result
-  }
-
-  const filteredContent = typedResult.content.filter((block) => block.type !== "image")
-
-  const textBlock = filteredContent.find((block) => block.type === "text")
-  let screenshotPath: string | null = null
-
-  if (textBlock?.text) {
-    // Extract file path from text like "saved it as /path/to/screenshot.png"
-    const pathMatch = textBlock.text.match(/(?:saved (?:it )?as|saved to) (.+\.png)/i)
-    if (pathMatch) {
-      screenshotPath = pathMatch[1]
-    }
-  }
-
-  // Add helpful message about reading the screenshot
-  if (screenshotPath && textBlock) {
-    filteredContent.push({
-      type: "text",
-      text: `\n\nTo view this screenshot, use the "read" tool with the file path: ${screenshotPath}`,
-    })
-  }
-
-  return { ...typedResult, content: filteredContent }
-}
-
 export async function handler(args: BrowserEvalArgs): Promise<string> {
   try {
     if (args.action === "start") {
@@ -327,13 +288,10 @@ export async function handler(args: BrowserEvalArgs): Promise<string> {
 
     const result = await callServerTool(connection, toolName, toolArgs)
 
-    const formattedResult =
-      args.action === "screenshot" ? filterImageDataFromResult(result) : result
-
     return JSON.stringify({
       success: true,
       action: args.action,
-      result: formattedResult,
+      result,
     })
   } catch (error) {
     const errorMessage = error instanceof Error ? error.message : String(error)

test/unit/browser-eval-filter.test.ts

@@ -0,0 +1,73 @@
+import { describe, it, expect, vi, beforeEach, afterEach } from "vitest"
+
+// Mock the mcp-client module before importing browser-eval-manager
+vi.mock("../../src/_internal/mcp-client.js", () => ({
+  connectToMCPServer: vi.fn(),
+  callServerTool: vi.fn(),
+  listServerTools: vi.fn(),
+}))
+
+// Mock the exec function to skip installation check
+vi.mock("child_process", () => ({
+  exec: vi.fn((cmd, callback) => {
+    callback(null, { stdout: "@playwright/[email protected]" })
+  }),
+}))
+
+import { connectToMCPServer, callServerTool } from "../../src/_internal/mcp-client.js"
+import {
+  startBrowserEvalMCP,
+  stopBrowserEvalMCP,
+} from "../../src/_internal/browser-eval-manager.js"
+import { handler } from "../../src/tools/browser-eval.js"
+
+describe("browser-eval playwright-mcp screenshot tool", () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+    vi.mocked(connectToMCPServer).mockResolvedValue({
+      client: { close: vi.fn() },
+      transport: { close: vi.fn() },
+    } as never)
+  })
+
+  afterEach(async () => {
+    await stopBrowserEvalMCP()
+  })
+
+  it("should pass --image-responses omit flag to playwright-mcp", async () => {
+    await startBrowserEvalMCP()
+
+    expect(connectToMCPServer).toHaveBeenCalledWith(
+      "npx",
+      expect.arrayContaining(["--image-responses", "omit"]),
+      expect.any(Object)
+    )
+  })
+
+  it("should return screenshot file path without base64 image data", async () => {
+    await startBrowserEvalMCP()
+
+    const screenshotPath = "/var/folders/tmp/screenshot-1234.png"
+    vi.mocked(callServerTool).mockResolvedValue({
+      content: [{ type: "text", text: `Screenshot saved to ${screenshotPath}` }],
+    })
+
+    const result = await handler({ action: "screenshot" })
+    const parsed = JSON.parse(result)
+
+    expect(parsed.success).toBe(true)
+    expect(parsed.action).toBe("screenshot")
+
+    // Verify response contains file path
+    const textContent = parsed.result.content.find(
+      (block: { type: string }) => block.type === "text"
+    )
+    expect(textContent.text).toContain(screenshotPath)
+
+    // Verify no base64 image data in response
+    const imageContent = parsed.result.content.find(
+      (block: { type: string }) => block.type === "image"
+    )
+    expect(imageContent).toBeUndefined()
+  })
+})