The Model Context Protocol (MCP) Python SDK versions 1.9.1 through 1.29.1 and 2.0.0 through 2.1.1 contain a vulnerability in their OAuth client implementation. The SDK does not properly validate the authorization server issuer metadata and does not bind stored client credentials to a specific authorization server. This allows a malicious or compromised MCP server to redirect OAuth credentials to an attacker-controlled authorization server, potentially exposing client secrets and authorization codes. The issue affects OAuth clients using certain providers and connecting to untrusted MCP servers. Fixed in versions 1.30.0 and 2.2.0, the fix includes strict issuer validation and credential binding. Users should upgrade and rotate credentials if exposure is suspected.
Mistune versions prior to 3.3.4 contain a quadratic-time denial-of-service vulnerability in the parse_link_href function. This occurs due to an unbounded parenthesis scan triggered by empty-label links, causing excessive CPU usage when rendering crafted markdown inputs. The issue is fixed in version 3.3.4.
Mistune versions prior to 3.3.4 contain a quadratic-time denial-of-service vulnerability in the inline image parsing logic. The vulnerability arises because certain guards intended to limit processing time are bypassed for image syntax, causing excessive CPU usage when parsing crafted markdown input with repeated image openers. This can cause a single unauthenticated request to consume 100% CPU on a worker, significantly degrading service availability. The issue is fixed in version 3.3.4.
A vulnerability in tiny-secp256k1 versions >=2.1.0 <2.2.5 and >=2.2.2 <2.2.5 allows an invalid tweak operation to be accepted due to a stale buffer comparison after a wasm failure. Specifically, the three-argument form of xOnlyPointAddTweakCheck ignores the wasm failure return and compares an unchanged input buffer, causing it to incorrectly validate an invalid derivation when the tweak results in the point at infinity. This flaw is fixed in version 2.2.5.
A vulnerability in fzf's zsh completion script allows local users on multi-user hosts to pre-create a FIFO path in /tmp as a symlink to victim-writable files, causing those files to be overwritten with attacker-influenced content. This can lead to arbitrary file overwrite and potential code execution at the victim's next login. The issue arises because the script uses a predictable FIFO path based on the shell PID and does not properly check the success of removing or creating the FIFO. Kernel protections such as fs.protected_symlinks may block exploitation on modern systems. A fix is available in version 0.74.5.
The fzf-tmux script in fzf versions prior to 0.74.5 improperly validates layout size arguments in attached form options (-l/-u), allowing bash arithmetic expansion to execute arbitrary commands. Additionally, unquoted tmux popup geometry options enable injection of tmux flags, leading to command execution via environment variable sourcing. This local vulnerability requires attacker-controlled input to layout arguments and affects the invoking user. A fix is available in version 0.74.5.
The npm package webhook-easy version 1.1.0 contains malicious code that executes arbitrary JavaScript fetched from a remote, attacker-controlled URL. This code decrypts and evaluates the fetched content at runtime, enabling remote code execution. The package also includes shadow dependencies that mimic Node.js core modules, potentially introducing additional malicious behavior. This activity is unrelated to the package's stated purpose and represents a supply chain compromise.
CVE-2026-133435 is a vulnerability in Red Hat Build of Keycloak where case-insensitive database collation causes realm URLs to be resolved regardless of letter casing. This allows remote attackers to bypass upstream reverse proxy access controls that depend on case-sensitive path matching, potentially exposing restricted internal authentication endpoints. The internal authentication mechanisms of Keycloak remain unaffected. The CVSS score is 3.7, indicating low severity.
sqlite-mcp versions prior to 0.1.2 allow bypassing read-only mode protections via statement chaining in execute_query and SQL injection in describe_table. This enables unauthorized write access to databases intended to be read-only, reading arbitrary SQLite files, and creating new SQLite files where the server process can write. The server lacks built-in authentication and listens on all interfaces by default, increasing exposure risk. The issue is fixed in version 0.1.2 by disabling ATTACH, enforcing query-only mode, rejecting multi-statement queries, and parameterizing inputs.
Next.js versions from 15.0.0 up to but not including 15.5.29, and from 16.0.0 up to but not including 16.4.5, have a vulnerability where HttpOnly cookies set via NextResponse on Edge runtime routes are exposed through an internal response header. This header duplicates the cookie value and attributes, making sensitive HttpOnly cookies accessible to JavaScript on the same origin. This issue affects App Router route handlers and Pages Router Edge API routes. A fix is available in versions 15.5.29 and 16.4.5.
A vulnerability in Next.js image optimizer allows processing of attacker-supplied SVG files even when SVG support is disabled by default. This bypasses the intended restriction and exposes the application to potential denial of service and SVG decoder vulnerabilities. The issue affects self-hosted deployments of Next.js versions 15.0.0 up to but not including 15.5.29, and 16.0.0 up to but not including 16.4.5. A patch is available in versions 15.5.29 and 16.4.5.
The npm package @cexy/wonfca, which claims to provide a Facebook Chat API, contains malicious code that routes user authentication credentials through a hardcoded third-party server (https://minhdong.site) controlled by the package author. When users invoke the package's login function, their Facebook email, password, and two-factor authentication data are sent via HTTP POST to this unauthorized server. This server then intermediates the login process, capturing live Facebook session tokens and cookies, effectively compromising user credentials and sessions.
A vulnerability in self-hosted Next.js applications using App Router route handlers and fetch cache can cause cross-client response injection and disclosure. This occurs when the application returns an upstream fetched response unchanged, leading to mismatched response framing. When deployed behind an HTTP/1.1 reverse proxy that allows pipelined response queueing, crafted upstream responses can desynchronize client responses, enabling injection and disclosure of other clients' data. Fixed versions 16.4.5 and 15.5.29 address this issue.
Next.js App Router applications using the `use cache` directive with binary arguments (e.g., Uint8Array or Blob) may generate identical cache keys for different inputs. This flaw can cause cached data to be shared across users, leading to cross-user cache disclosure or cache poisoning. The vulnerability affects Next.js versions from 16.0.0 up to but not including 16.4.5. A fix is available in version 16.4.5.
Next.js versions 16.4.0 through 16.4.4, when self-hosted with cacheComponents enabled and partialPrefetching disabled, are vulnerable to cache poisoning of shared fallback shells via request-supplied placeholder route parameters. An unauthenticated attacker can poison the fallback shell cache entry shared by a family of routes, causing persistent 404 responses or incorrect parameter rendering in server-side HTML. No script injection or data disclosure occurs. The issue is fixed in version 16.4.5.
A JavaScript injection vulnerability exists in safari-mcp versions prior to 2.22.15 due to improper escaping of backslashes in single-quoted JavaScript string literals in multiple tool arguments. This flaw allows an attacker to inject arbitrary JavaScript code in the page context, including in tabs not owned by the session, even when limited to read-only tools. The issue is fixed in version 2.22.15 by properly escaping backslashes before quotes.
CVE-2026-133494 is a vulnerability in Greenbone gvmd versions prior to 26.43.0 where certificate creation is not properly restricted. This allows even users with the Observer role, which is intended to be read-only, to create TLS certificates. The issue is fixed in version 26.43.0.
CVE-2026-133430 is a vulnerability in Red Hat Build of Keycloak where improper verification of cryptographic signatures in SAML logout requests allows an authenticated attacker to terminate active sessions of arbitrary users. The flaw arises because Keycloak accepts unsigned logout requests if any assertion within the SAML document is signed, without verifying the assertion's context or placement. This enables session termination denial-of-service attacks.
The certify-web package on PyPI version 0.1.3 contains malicious code that installs a PTH file. This file triggers on every Python startup to install a NinjaOne agent linked to an attacker-controlled account, enabling persistent remote control of the affected machine.
CVE-2026-115419 is a vulnerability in the Strands Agents SDK's Agent2Agent (A2A) server component that allows authenticated users to access conversation states of other users. This flaw affects multi-tenant deployments by compromising data isolation, enabling unauthorized access to conversation history or influencing model behavior across tenants. The issue impacts Python versions with the multiagent.a2a module prior to 1.43.0 and TypeScript SDK versions prior to 1.5.0. Fixed versions are 1.43.0 for Python and 1.5.0 for TypeScript.