CVE-2026-75612 is a path traversal vulnerability in yt-mcp, a YouTrack MCP server for Claude Code, affecting versions prior to 1.18.1. The add_attachment function improperly handles a caller-controlled file_path, allowing reading of arbitrary files accessible to the server process. Additionally, get_project_health allows directory traversal to write JSON files outside the intended snapshots directory. These issues enable unauthorized file read and write operations. The vulnerability is fixed in version 1.18.1.
Todoist MCP Server provides tools that allow AI agents to access and modify a Todoist account. Prior to 10.3.3, the optional todoist-mcp-http server accepts browser-origin requests to POST /mcp without validating the Host and Origin headers or authenticating the browser caller. A malicious website can use DNS rebinding to reach the loopback listener and invoke MCP tools through StreamableHTTPServerTransport, which uses the server-held Todoist API key. This exposes Todoist account data and permits operations such as task creation, updates, completion, and deletion. The stdio transport is not browser-reachable and is not affected by this attack path. This issue is fixed in version 10.3.3.
Obot versions from 0.14.0 up to but not including 0.27.0 have a server-side request forgery (SSRF) vulnerability. This flaw allows privileged users to bypass outbound network protections by using certain unspecified addresses, IPv6 transition addresses, or registry icon URLs. Attackers can exploit this to add remote MCP servers pointing to special IP addresses such as 0.0.0.0, ::, or NAT64/6to4/Teredo addresses that embed blocked IPv4 addresses, enabling probing of internal services for reachability.
CVE-2026-130767 is a command injection vulnerability in Obot versions from 0.9.0 up to but not including 0.27.0. Authenticated users can exploit this flaw by embedding an EOF line in file-type MCP server configuration values, allowing them to terminate heredoc in the init container script and execute arbitrary shell commands. This can lead to modification of MCP server files and access to services on Obot's Docker network. The vulnerability has a CVSS 4.0 score of 5.3, indicating medium severity. A fix is available in version 0.27.0.
CVE-2026-130764 is a server-side request forgery (SSRF) vulnerability in obot-platform's obot product versions from 0.8.0 up to but not including 0.27.0. This flaw allows Power User accounts to exploit malicious OAuth token endpoints advertised in remote MCP server metadata to reach internal hosts. Attackers can manipulate token exchange or refresh requests, including redirected GET requests, to access cloud metadata services or private network hosts and read error response bodies.
sqlite-mcp is a Model Context Protocol server for SQLite. Prior to 0.1.2, the default read-only mode can be bypassed through execute_query, which accepts a leading SELECT while the SQLite driver executes trailing statements, and through describe_table, which interpolates the caller-controlled table_name into PRAGMA table_info. Either path can issue ATTACH DATABASE, which opens an attached file read-write despite mode=ro and can make the protected database writable, read other SQLite database files, or create a SQLite-format file at a path writable by the server process. Existing non-empty files that are not SQLite databases are refused and left unchanged. The server has no built-in authentication and listens on all interfaces at :8080 by default, so any client that can reach the port can invoke these tools unless an authenticating proxy is in place. This issue is fixed in version 0.1.2.
CVE-2026-130007 is a vulnerability in k8s-mcp-server prior to version 1.1.0 where critical Kubernetes management functionality lacks authentication. The server listens on all interfaces by default and allows unauthenticated clients to enumerate and invoke Kubernetes tools with the server's cluster privileges. This can lead to unauthorized execution of write-capable operations. The issue is fixed in version 1.1.0.
Supergateway versions 2.4.0 through 4.2.2 have a cross-site WebSocket hijacking vulnerability due to missing Origin header validation in stdio-to-WebSocket mode. This flaw allows malicious web pages to connect to the /message WebSocket endpoint from a victim's browser, including localhost-bound gateways, enabling attackers to initiate MCP sessions and invoke wrapped stdio MCP server tools with the victim's privileges.
The AI Puffer - MCP Server for Claude & ChatGPT, with AI Chatbot plugin for WordPress is vulnerable to Stored Cross-Site Scripting via AI-generated comment replies in all versions up to, and including, 2.5.28 due to insufficient sanitization of the AI model's output in the insert_comment_reply_logic() function, which saves replies with wp_insert_comment() and bypasses the kses filtering WordPress normally applies to comment content. This makes it possible for unauthenticated attackers to inject arbitrary web scripts in pages that will execute whenever a user accesses an injected page, by submitting a comment that leads the AI model to reproduce an HTML payload in its automated reply. This is only exploitable when an administrator has configured a Comment Auto-Reply automation task with 'Action on reply' set to 'Approve' (the default is 'Hold for moderation'), and the attacker's comment has been approved.
Flowise versions 3.1.3 through 3.1.4 have an information disclosure vulnerability that permits unauthenticated attackers to access MCP server bearer tokens via a public API endpoint. These tokens can then be used to bypass API key protections and invoke chatflows, consuming the owner's model quota.
The mcp-aevo-server package on PyPI is a malicious package that targets users of ribbon-finance/aevo-trading-skills by exfiltrating environment variable credentials and executing remote commands via a listening thread. It employs dependency confusion techniques and uses a Telegram bot for command and control.
code-review-mcp-server automates code review and can post comments directly on GitHub pull requests. Prior to 1.5.3, the getLocalGitDiff, codeReview, createPR, and getPRTemplate MCP tools accept caller-controlled folderPath and baseBranch string parameters that are interpolated into command strings executed by child_process.execSync through /bin/sh. Shell metacharacters, command substitution, or quote-breaking input can therefore execute arbitrary commands with the privileges of the host process running the MCP server. This issue is fixed in version 1.5.3.
Improper restriction of intended endpoints in the connection configuration handling of the AWS Labs Timestream for InfluxDB MCP Server (awslabs.timestream-for-influxdb-mcp-server) before 0.0.19 might allow a remote unauthenticated actor (via prompt injection) to obtain the InfluxDB API token configured on the server, sent to a crafted endpoint controlled through an InfluxDB URL introduced in the MCP client context.
To remediate this issue, users should upgrade to version 0.0.19 or later and rotate any InfluxDB API tokens configured for the server.
The Model Context Protocol Python SDK is the official Python SDK for Model Context Protocol servers and clients. From 1.9.1 until 1.30.0 and 2.2.0, OAuth client support in mcp.client.auth can send credentials to an authorization server chosen by an untrusted MCP server. ClientCredentialsOAuthProvider and PrivateKeyJWTOAuthProvider cannot identify the authorization server that owns their credentials. Affected HTTP clients using those providers, OAuthClientProvider, or the deprecated 1.x RFC7523OAuthClientProvider can disclose a client_secret, authorization code, PKCE code_verifier, or signed client assertion. MCP servers, stdio clients, and clients that attach their own tokens or headers are not affected. Interactive OAuthClientProvider use requires a person to start sign-in, while the unattended providers do not. For ClientCredentialsOAuthProvider and PrivateKeyJWTOAuthProvider, callers must configure issuer= after upgrading because omission still follows the MCP server's advertised authorization server, and the deprecated RFC7523OAuthClientProvider must be replaced because it has no issuer option. A stored client registration without an issuer remains unbound after upgrading, so stored OAuth client information must be cleared once to register again or a manually pre-registered record must be updated with its issuer. This issue is fixed in versions 1.30.0 and 2.2.0.
Improper limitation of a pathname to a restricted directory in the cost report generation tool in AWS Pricing MCP Server (aws-pricing-mcp-server) before 1.0.31 might allow an actor who can influence the MCP agent to write actor controlled content to arbitrary locations outside the intended output directory via absolute paths or directory traversal sequences in the report output path.
To remediate this issue, users should upgrade to version 1.0.31 or later. Users of the deprecated awslabs.cost-analysis-mcp-server package should migrate to awslabs.aws-pricing-mcp-server version 1.0.31 or later.