CVEReports
CVEReports

Automated vulnerability intelligence platform. Comprehensive reports for high-severity CVEs generated by AI.

Product

  • Home
  • Sitemap
  • RSS Feed

Company

  • About
  • Contact
  • Privacy Policy
  • Terms of Service

© 2026 CVEReports. All rights reserved.

Made with love by Amit Schendel & Alon Barad



CVE-2026-75518

CVE-2026-75518: Read-Tier Authorization Bypass in DOMShell MCP Server

Amit Schendel
Amit Schendel
Senior Security Researcher

Nov 19, 2026·4 min read·1 visit

Executive Summary (TL;DR)

Prior to version 2.0.8, the @apireno/domshell MCP server misclassified the domshell_eval tool and eval command as read-only operations. Because the backend evaluates these expressions via Chrome DevTools Protocol Runtime.evaluate without side-effect restrictions, an MCP caller lacking --allow-write privileges can execute arbitrary JavaScript to mutate DOM, window, and storage state.

CVE-2026-75518 is an incorrect authorization and privilege assignment vulnerability in the DOMShell Model Context Protocol (MCP) server (@apireno/domshell) prior to version 2.0.8. The server classified the domshell_eval tool and the eval command as read-tier operations, permitting their execution without the --allow-write flag. Because the eval command dispatches through the same Chrome DevTools Protocol (CDP) Runtime.evaluate pipeline as the write-tier js command without enabling the throwOnSideEffect constraint, callers restricted to read-only sessions can execute state-mutating JavaScript in the browser context while host user interfaces and audit logs record the activity as read-only.

Vulnerability Overview

DOMShell (@apireno/domshell) is a Google Chrome automation platform that exposes Unix-style filesystem and interaction commands to artificial intelligence agents via a Model Context Protocol (MCP) server. The architecture consists of a Node.js MCP server wrapper (mcp-server/index.ts) that communicates over a local WebSocket bridge with a companion Chrome extension. The extension translates incoming commands into Chrome DevTools Protocol (CDP) instructions executed against browser tabs.

To enforce least-privilege access for automated agents, the DOMShell MCP server divides commands into distinct authorization tiers: read, navigate, write, and sensitive. By default, the MCP server operates in a restricted read-only mode unless the operator explicitly passes the --allow-write command-line flag. Commands in the read tier (such as ls, cat, text, tree, and find) are intended for non-destructive DOM inspection, whereas commands in the write tier (such as click, type, select, and js) modify page state and require explicit write authorization.

In versions of @apireno/domshell prior to 2.0.8, the MCP server failed to include the eval command in the write-tier command set (WRITE_COMMANDS) and annotated the domshell_eval MCP tool with read-only metadata (ANNO_READ). This defect represents a combination of Incorrect Authorization (CWE-863) and Incorrect Privilege Assignment (CWE-266).

Although the tool documentation and metadata declared domshell_eval to be a read-only expression evaluator, the backend extension routed both eval and js to the exact same CDP Runtime.evaluate execution handler without enabling side-effect protection. Consequently, any MCP client or AI agent operating in a read-only session could invoke domshell_eval or domshell_execute with an eval subcommand to run arbitrary, state-mutating JavaScript in the context of the target web page.

Root Cause Analysis

The root cause of CVE-2026-75518 lies in an architectural mismatch between the command authorization policy enforced by the MCP server (mcp-server/index.ts) and the JavaScript evaluation implementation inside the Chrome extension background script (src/background/index.ts and src/background/cdp_client.ts).

In the MCP server layer, authorization checks determine a command's security tier by checking its name against static Set definitions (NAVIGATE_COMMANDS, WRITE_COMMANDS, and SENSITIVE_COMMANDS). Any command not present in these sets defaults to the read tier and is allowed to execute without the --allow-write flag. Prior to version 2.0.8, WRITE_COMMANDS included the js command but omitted eval. Furthermore, the domshell_eval tool registration passed the ANNO_READ annotation (readOnlyHint: true) to the MCP SDK.

When the MCP server forwards an authorized command across the WebSocket bridge to the Chrome extension, the extension does not perform secondary tier validation. In src/background/index.ts (lines 1849–1854), the command router maps both eval and js directly to the handleJs function, which subsequently invokes cdp.evaluateJs in src/background/cdp_client.ts (lines 660–667).

The cdp.evaluateJs method dispatches a CDP Runtime.evaluate request to the V8 engine with the parameters expression: code, returnByValue: true, and awaitPromise: true. The Chrome DevTools Protocol supports a boolean parameter named throwOnSideEffect, which instructs V8 to abort execution if the evaluated script attempts to modify state or trigger DOM mutations. Because cdp.evaluateJs does not set throwOnSideEffect: true, V8 executes the expression with full read/write capabilities.

This design created two compound failures. First, the server authorization gate allowed unprivileged sessions to reach an unrestricted JavaScript execution sink. Second, the ANNO_READ tool annotation instructed MCP host applications (such as Claude Desktop or Cursor) to display the operation to human operators as a safe, read-only query in approval prompts and recorded it as a read-tier action in server audit logs.

Code Analysis

The vulnerability was resolved in commit 03c1d7625e220297aff0dde37613d65cfe6f9a36 by reclassifying eval as a write-tier command within the MCP server and updating its MCP tool annotation and system prompt documentation.

In mcp-server/index.ts, the vulnerable code defined WRITE_COMMANDS without `

Technical Appendix

CVSS Score
4.8/ 10
CVSS:4.0/AV:L/AC:L/AT:N/PR:L/UI:N/VC:L/VI:L/VA:N/SC:N/SI:N/SA:N
EPSS Probability
0.11%
Top 99% most exploited
CWE-863
Incorrect Authorization

The software performs an authorization check when an actor attempts to access a resource or perform an action, but does not correctly determine whether that actor is authorized to perform that action.

More Reports

•about 2 hours ago•CVE-2026-63411
9.3

CVE-2026-63411: Remote Code Execution via PHPDoc Comment Breakout in Jane PHP OpenAPI Generator

CVE-2026-63411 (GHSA-xwcw-j52r-34x4) is a remote code execution vulnerability in the Jane PHP code generation framework (janephp/janephp), specifically affecting the jane-php/open-api-3 component prior to version 7.12.0. The generator constructs PHP docblocks using unsanitized parameter names and descriptions extracted directly from OpenAPI specifications. By embedding the comment closing sequence (*/) followed by valid PHP syntax inside an OpenAPI description or parameter name, an attacker can break out of the PHPDoc context and inject executable code directly into the generated class body. When the consuming application instantiates the generated client or endpoint class, an injected __construct() method executes arbitrary PHP code under the privileges of the host runtime.

Amit Schendel
Amit Schendel
2 views•9 min read
•about 3 hours ago•CVE-2026-63413
9.3

CVE-2026-63413: Remote Code Execution via PHPDoc Comment Breakout in Jane PHP OpenAPI 3 Generator

CVE-2026-63413 is a code injection vulnerability in the Jane PHP OpenAPI 3 SDK generator (jane-php/open-api-3) prior to version 7.12.0. The vulnerability occurs because the generator interpolates OpenAPI header and query parameter names directly into generated PHPDoc comment blocks without sanitizing comment closing delimiters (*/). An attacker who supplies or modifies an OpenAPI specification can break out of the docblock context and inject arbitrary PHP methods, such as a class constructor, into the generated Client or Endpoint class body. When the consuming application instantiates the generated SDK, the injected code executes with the privileges of the host PHP process.

Alon Barad
Alon Barad
2 views•9 min read
•about 4 hours ago•CVE-2026-63412
9.3

CVE-2026-63412: Remote Code Execution via PHPDoc Comment Injection in Jane PHP OpenAPI Generator

CVE-2026-63412 is a code injection vulnerability in the Jane PHP library suite (affecting jane-php/open-api-3, jane-php/open-api-2, jane-php/open-api-31, and jane-php/json-schema) prior to version 7.12.0. The flaw occurs during the automated generation of PHP client SDKs and data models from OpenAPI specifications. While Jane uses an Abstract Syntax Tree (AST) for safe generation of standard PHP tokens, it constructs PHPDoc comments via raw string interpolation without neutralizing multi-line comment closing delimiters (*/). An attacker who provides or modifies an OpenAPI specification can inject a closing comment sequence inside a query parameter name, escaping the docblock context and writing arbitrary PHP code directly into the class body of generated endpoints or clients. When the consuming application instantiates the generated class, injected methods such as __construct() execute automatically, resulting in remote code execution.

Amit Schendel
Amit Schendel
3 views•9 min read
•about 5 hours ago•CVE-2026-58214
4.3

CVE-2026-58214: MQTT Internal Subject Subscription ACL Bypass in NATS Server

CVE-2026-58214 is an incorrect authorization vulnerability (CWE-863) in the MQTT bridge implementation of NATS Server. The vulnerability results from an incomplete fix for CVE-2026-33217 within the subscription processing routine in server/mqtt.go. Because internal delivery subscriptions bypass standard account-level allow-list validation in canSubscribe, the MQTT session manager relies on explicit prefix filtering in processSubs to prevent clients from subscribing to internal broker subjects. Affected versions fail to validate incoming MQTT subscription filters against the internal QoS 2 PubRel delivery prefix ($MQTT.deliver.pubrel.), allowing authenticated MQTT clients to subscribe to internal operational subjects and monitor QoS 2 protocol metadata across all sessions within the same NATS account.

Amit Schendel
Amit Schendel
5 views•9 min read
•about 7 hours ago•CVE-2026-69155
8.7

CVE-2026-69155: Stored Cross-Site Scripting in Atomic Server File Download Handler

CVE-2026-69155 (GHSA-x277-3wcg-g9r2) is a stored cross-site scripting (XSS) vulnerability in Atomic Server versions 0.30.0 prior to 0.40.3. The vulnerability resides in the HTTP file-download handler (server/src/handlers/download.rs), which serves user-uploaded files with an inline Content-Disposition and attacker-controlled MIME types such as text/html or image/svg+xml. Because the endpoint lacks X-Content-Type-Options: nosniff and Content-Security-Policy restrictions, an authenticated user with upload privileges can store malicious JavaScript that executes within a victim's browser in the primary application origin. Executing code in this context grants access to non-HttpOnly session cookies and unencrypted Ed25519 cryptographic private keys stored in IndexedDB, leading to full account takeover.

Amit Schendel
Amit Schendel
4 views•10 min read
•about 8 hours ago•CVE-2026-69156
8.1

CVE-2026-69156: Arbitrary File Read and Root Key Disclosure via Path Traversal in Atomic Server

CVE-2026-69156 is an authenticated path traversal and arbitrary file read vulnerability in Ontola Atomic Server versions 0.30.0 prior to 0.40.3. The flaw arises from a failure to restrict server-managed properties during Commit ingestion in lib/src/commit.rs, combined with unsanitized filesystem path construction in the /download endpoint handler in server/src/handlers/download.rs. An authenticated Agent can assign an arbitrary relative or absolute filesystem path to the internalId property of an owned resource and retrieve the target file via the download endpoint, exposing the server configuration file (config.toml) and the root Agent ED25519 private key.

Alon Barad
Alon Barad
7 views•12 min read