Claude-generated logging code wrote upload credentials and signed request URLs to debug output. Anyone able to read the logs could recover sensitive tokens and request data.

How AI contributed: Direct introductionPublished Mar 9, 2026 in lupinlin1/jimeng-web-mcp

Show full advisory description

The Claude-generated release converted the console logging to the project logger and added new debug lines that record the upload credential response in ImageUploader.getUploadAuth() and the fully expanded request URL in HttpClient.request(). The logged URL carries signed query parameters and cookie-derived tokens, and the logger masks only the context keys it recognises, so neither value is redacted. Anyone with access to the server logs can recover live upload credentials and signed request links. No fix has been released.

What actually happened

The sensitive-data logging entered with the AI-written change released in v2.1.2, the only version inside the affected range, and persisted for that entire version.

Root cause

Authentication & access control introduced

fix: repair MCP stdio communication and unresponsive image uploads

AI-assisted change: Claude

eba7d2efe0

src/api/ImageUploader.ts src/api/HttpClient.ts src/server.ts

Fix status

No public fix identified

No minimum fix commit has been established for this finding.

Fix status unresolved

Case facts

CVE ID: CVE-2025-70040
GHSA ID: GHSA-V9V4-F5WM-PHH4
Published: Mar 9, 2026
Severity: Not scored
Language: TypeScript
AI tool: Claude
AI role: Direct introduction
Vulnerable release: = 2.1.2

Advisory source

AI-assisted change

AI-assisted changesrc/api/ImageUploader.ts source lupinlin1/jimeng-web-mcp@eba7d2efe0+27−11

@@ -13,6 +13,7 @@ import path from 'path'; import crc32 from 'crc32'; import { HttpClient } from './HttpClient.js'; import { retryAsync } from '../utils/retry.js';+import { logger } from '../utils/logger.js';  export interface UploadResult {   uri: string;           // 服务器返回的URL

Show 82 hidden lines

@@ -39,19 +40,25 @@ export class ImageUploader {    * 上传单张图片    */   async upload(imagePath: string): Promise<UploadResult> {+    let uploadAuth;     try {       // 获取上传凭证-      const uploadAuth = await this.getUploadAuth();+      uploadAuth = await this.getUploadAuth();+    } catch (error) {+      throw new Error(`图片上传失败 [${imagePath}] at 步骤1 (获取上传凭证): ${error}`);+    } -      // 读取图片内容-      const imageBuffer = await this.getFileContent(imagePath);+    // 读取图片内容+    const imageBuffer = await this.getFileContent(imagePath); -      // 使用image-size库检测格式和尺寸（替代132行手动解析）-      const metadata = this.detectFormat(imageBuffer);+    // 使用image-size库检测格式和尺寸（替代132行手动解析）+    const metadata = this.detectFormat(imageBuffer); -      // 计算CRC32校验-      const imageCrc32 = crc32(imageBuffer).toString(16);+    // 计算CRC32校验+    const imageCrc32 = crc32(imageBuffer).toString(16); +    let uploadImgRes;+    try {       // 准备上传参数       const getUploadImageProofRequestParams = {         Action: 'ApplyImageUpload',@@ -75,7 +82,7 @@ export class ImageUploader {       const getUploadImageProofUrl = 'https://imagex.bytedanceapi.com/';        // 获取图片上传凭证-      const uploadImgRes = await this.httpClient.request({+      uploadImgRes = await this.httpClient.request({         method: 'GET',         url: getUploadImageProofUrl + '?' + this.httpClient.httpBuildQuery(getUploadImageProofRequestParams),         headers: requestHeadersInfo@@ -84,20 +91,29 @@ export class ImageUploader {       if (uploadImgRes?.['Response']?.hasOwnProperty('Error')) {         throw new Error(uploadImgRes['Response']['Error']['Message']);       }+    } catch (error) {+      throw new Error(`图片上传失败 [${imagePath}] at 步骤2 (获取图片上传凭证): ${error}`);+    } -      const UploadAddress = uploadImgRes.Result.UploadAddress;+    const UploadAddress = uploadImgRes.Result.UploadAddress; +    try {       // 上传图片       const uploadImgUrl = `https://${UploadAddress.UploadHosts[0]}/upload/v1/${UploadAddress.StoreInfos[0].StoreUri}`;        // 图片上传步骤：带重试机制（仅此步骤重试）-      const imageUploadRes = await this.uploadImageDataWithRetry(+      await this.uploadImageDataWithRetry(         uploadImgUrl,         imageBuffer,         imageCrc32,         UploadAddress.StoreInfos[0].Auth       );+    } catch (error) {+      throw new Error(`图片上传失败 [${imagePath}] at 步骤3 (上传图片数据): ${error}`);+    } +    let commitRes;+    try {       // 提交上传       const commitImgParams = {         Action: 'CommitImageUpload',@@ -123,7 +139,7 @@ export class ImageUploader {        const commitImgUrl = 'https://imagex.bytedanceapi.com/'; -      const commitRes = await this.httpClient.request({+      commitRes = await this.httpClient.request({         method: 'POST',         url: commitImgUrl + '?' + this.httpClient.httpBuildQuery(commitImgParams),

Lifts upload authentication, upload response, and commit result values into a shared scope while importing the debug logger, allowing their sensitive states to be written outside the local try blocks.

AI-assisted changesrc/api/HttpClient.ts source lupinlin1/jimeng-web-mcp@eba7d2efe0+6−2

@@ -9,6 +9,7 @@ import { generateCookie } from '../utils/auth.js'; import { DEFAULT_ASSISTANT_ID, UA } from '../types/models.js'; import { generateUuid, jsonEncode, toUrlParams, generateMsToken, unixTimestamp } from '../utils/index.js'; import { generate_a_bogus } from '../utils/a_bogus.js';+import { logger } from '../utils/logger.js'; import crypto from 'crypto';  export interface RequestOptions {

Show 35 hidden lines

@@ -50,7 +51,7 @@ export class HttpClient {       data = {},       headers = {},       params = {},-      timeout = 60000+      timeout = 120000 // 增加超时时间到120秒     } = options;      const baseUrl = 'https://jimeng.jianying.com';@@ -61,6 +62,7 @@ export class HttpClient {       "Accept-Encoding": "gzip, deflate, br, zstd",       "Accept-language": "zh-CN,zh;q=0.9",       "Cache-control": "no-cache",+      "Content-Type": "application/json",  // 🔥 添加Content-Type       "Last-event-id": "undefined",       Appid: DEFAULT_ASSISTANT_ID,       Appvr: "5.8.0",@@ -84,6 +86,8 @@ export class HttpClient {       ...headers     }; +    logger.debug(`[HttpClient] Request: ${method} ${fullUrl}`);+     try {       const response: AxiosResponse<T> = await axios({         method: method.toLowerCase(),@@ -302,7 +306,7 @@ export class HttpClient {     if (error.response) {       throw new Error(`即梦API请求错误: ${JSON.stringify(error.response.data)}`);     } else if (error.request) {-      throw new Error(`即梦API请求失败: 无响应`);+      throw new Error(`[FINAL-DEBUG] HttpClient.handleError: Caught error with no response.`);     } else {       throw new Error(`即梦API请求失败: ${error.message}`);     }

The HTTP client now logs the outgoing request method and fully expanded URL before dispatch, so signed query parameters and credential-bearing upload addresses become visible to debug output.

AI-assisted changesrc/server.ts source lupinlin1/jimeng-web-mcp@eba7d2efe0+8−8

@@ -126,23 +126,23 @@ export const createServer = (): McpServer => {           }]         };       } catch (error) {-        // 🔍 Debug logging - 记录详细错误信息-        console.error('🔍 [MCP Server] Error caught in generateImage tool:');-        console.error('🔍 [MCP Server] Error type:', error?.constructor?.name);-        console.error('🔍 [MCP Server] Error message:', error instanceof Error ? error.message : String(error));

Show 24 hidden lines

+        // 🔍 Debug logging - 记录详细错误信息 (使用logger以避免破坏stdio)+        logger.debug('🔍 [MCP Server] Error caught in generateImage tool');+        logger.debug(`🔍 [MCP Server] Error type: ${error?.constructor?.name}`);+        logger.debug(`🔍 [MCP Server] Error message: ${error instanceof Error ? error.message : String(error)}`);         if (error instanceof Error && error.stack) {-          console.error('🔍 [MCP Server] Error stack:', error.stack);+          logger.debug(`🔍 [MCP Server] Error stack: ${error.stack}`);         }-        +         // 🔍 记录错误时的参数状态-        console.error('🔍 [MCP Server] Parameters when error occurred:', JSON.stringify({+        logger.debug(`🔍 [MCP Server] Parameters when error occurred: ${JSON.stringify({           filePath: params.filePath,           prompt: params.prompt ? `${params.prompt.substring(0, 100)}...` : undefined,           model: params.model,           aspectRatio: params.aspectRatio,           sample_strength: params.sample_strength,           negative_prompt: params.negative_prompt-        }, null, 2));+        }, null, 2)}`);                  const errorMessage = error instanceof Error ? error.message : String(error);         return {

The generateImage error handler routes image-tool failure and parameter details into debug logging, capturing stack traces alongside session-specific prompt and path inputs.