A vulnerability in @modelcontextprotocol/server versions 2.0.0 through 2.2.0 allows an MCP server accessed over HTTP to deliver one client's responses and notifications to another client when clients share a stateless transport or a single server instance. This occurs because messages are matched only by JSON-RPC id, which can collide. The issue is fixed in version 2.3.0 and later, but requires code changes to avoid sharing SDK clients or server instances across multiple clients.
A vulnerability in the MCP TypeScript SDK and server allows a single HTTP request within the default 4 MiB size limit to exhaust server memory during validation by sending excessive array elements or object keys. This causes the server process to abort with a 'JavaScript heap out of memory' error. No authentication is required unless the server enforces it before the SDK processes requests. The issue affects versions 1.3.0 through 1.31.0 of @modelcontextprotocol/sdk and 2.0.0 through 2.2.0 of @modelcontextprotocol/server. Fixed versions are 1.32.0+ for the SDK and 2.3.0+ for the server, which introduce a new option to limit input elements and recommend lowering request size limits.
The MCP TypeScript SDK versions prior to specified patched releases accept bearer tokens issued for other services due to lack of audience (aud) validation in server-side authentication. This allows tokens valid for one service to be accepted by another MCP server using the SDK, if the authorization server issues tokens for multiple services. The issue affects multiple MCP SDK and server packages in versions before 1.32.0, 2.3.0, 2.0.2, and 2.3.1 respectively. A fix is available that adds an optional expectedResource setting to enforce audience validation, which must be enabled explicitly. Without enabling this setting or applying a custom audience check, the vulnerability remains exploitable.
The rmcp Streamable HTTP server transport versions 3.0.0 through 3.5.x have an unbounded cache of tool schema results keyed by tool name. This cache grows with each incoming tools/call request, including those rejected with HTTP 400, potentially leading to memory exhaustion. The vulnerability affects servers using protocol version 2026-07-28 or later and lacking authentication in front of the endpoint. The issue was fixed in version 3.6.0 by removing the cache and calling ServerHandler::get_tool on each request. Mitigation includes upgrading to 3.6.0 or implementing authentication or request filtering to block unknown tool names.
CVE-2026-113378 is an OS command injection vulnerability in the mcollina mcp-ripgrep project up to version 0.4.0. The flaw exists in the Advanced Search Tool component, specifically in the handling of the fileType argument in src/index.ts. The vulnerability allows local attackers to inject OS commands due to improper escaping of this argument, while sibling arguments are properly escaped. The exploit code has been publicly released, but the actual existence of the vulnerability is still questioned by some. The project maintainer warns that the MCP server executes shell commands using ripgrep and advises caution with input.
IBM ContextForge MCP Gateway 0.1.0 through 1.0.5 is vulnerable to a server-side request Forgery (SSRF) that allows authenticated users with gateway creation privileges to perform arbitrary HTTP POST requests to internal network resources and cloud metadata endpoints. The vulnerability occurs during OAuth-enabled gateway registration when the system fails to validate the oauth_config.token_url parameter. When registering a gateway with OAuth client credentials authentication, an attacker can supply a malicious token_url pointing to internal services, cloud metadata endpoints (AWS IMDSv1, GCP, Azure), or localhost-bound services. The application immediately performs an HTTP POST request to the attacker-controlled URL with form-encoded credentials in the body. If the response cannot be parsed as a valid OAuth token response, up to 256 characters of the response body are reflected back to the attacker in the HTTP 502 error message. This vulnerability bypasses existing SSRF protections that are applied to the gateway.url field but not to OAuth configuration parameters.
Bulletin ID: 2026-054-AWS Scope: AWS Content Type: Important (requires attention) Publication Date: 07/14/2026 13:00 PM PDT Description: AWS HealthLake MCP Server (awslabs.healthlake-mcp-server) is a Model Context Protocol server that enables AI assistants to interact with AWS HealthLake FHIR datastores. We identified CVE-2026-15643, a server-side request forgery in the pagination handling component in AWS awslabs.healthlake-mcp-server before 0.0.14 on all platforms might allow a remote authenticated user to exfiltrate AWS temporary security credentials to an arbitrary endpoint via a crafted next_token parameter. The server does not validate that pagination URLs point back to the expected HealthLake endpoint, allowing an actor to redirect subsequent requests to an actor-controlled server. Impacted versions: < 0.0.14 Please refer to the article below for the most up-to-date and complete information related to this AWS Security Bulletin.
CVE Intel is an open-source tool that aggregates multiple public vulnerability intelligence sources such as NVD, CISA KEV, EPSS, and OSV into a single risk score and prioritization recommendation. It supports querying CVEs, IPs, file hashes, and dependencies, providing a risk score from 0 to 100 and an urgency indicator for patching. The tool includes a local AI model for explaining findings and can be used as a desktop app, CLI, REST API, or MCP server for AI coding assistants. It does not scan or fix vulnerabilities but helps security teams prioritize patching efforts based on consolidated threat intelligence.
The CKAN MCP server's DataStore Table UI resource fails to validate the origin of inbound postMessage events and sends outbound messages with a wildcard target origin '*'. This allows any window or frame that can reach the document to inject arbitrary data into the table, causing content spoofing and interference with the UI's handshake process. The vulnerability does not allow script injection but can cause display spoofing and disrupt user experience. The issue is considered low severity due to its dependency on the embedding context and limited direct remote exploitability.
CVE-2026-111091 is a high-severity vulnerability in JetBrains Junie CLI versions before 26.8.3. It allows code execution through an untrusted project's MCP server configuration. The vulnerability is identified as CWE-829, which involves inclusion of untrusted code. The CVSS 3.1 base score is 7.8, indicating significant impact on confidentiality, integrity, and availability. No known exploits are reported in the wild. The issue is fixed in version 26.8.3.
n8n versions before 2.41.4 and 2.42.x before 2.42.1 have a prototype pollution vulnerability in the MCP workflow-validation tool. This flaw allows an authenticated user with only workflow:read permission to manipulate shared prototypes, bypassing server-side permission checks and potentially taking over owner accounts without MFA protection.
CVE-2026-108395 is a high-severity vulnerability in the Model Context Protocol Java SDK affecting multiple versions of the io.modelcontextprotocol.sdk:mcp artifact and mcp-core. The issue arises because certain server transports read HTTP request bodies into memory without enforcing a maximum size limit. This allows an unauthenticated remote attacker to send arbitrarily large POST requests, potentially exhausting the Java process heap and causing denial of service. The vulnerability affects versions from 0.8.0 up to but not including 0.13.0, from 0.13.0 up to 0.18.4, from 1.0.0 up to 1.1.4, and from 2.0.0-M1 up to 2.0.1. Fixed versions are 0.18.4, 1.1.4, and 2.0.1. Mitigations include authentication, origin or host validation, or upstream request size limits to reject malicious requests before they reach the vulnerable code.
CVE-2026-109402 is a server-side request forgery (SSRF) vulnerability in conductor-oss conductor version 3.22.0. It affects the MCPService.listTools function in the MCP Service component. The vulnerability allows remote attackers to manipulate the mcpServer argument to cause SSRF. The vulnerability has a medium severity with a CVSS 4.0 score of 5.3. Exploit code has been publicly disclosed, but no known active exploitation in the wild has been reported. No patch or remediation information is currently available.
CVE-2026-77363 is a high-severity vulnerability in obsidian-tc, a Model Context Protocol server for Obsidian vaults. Versions prior to 1.3.1 do not validate the Host or Origin HTTP headers and lack DNS-rebinding protection in the Streamable HTTP transport. When configured with auth.mode set to none in the documented loopback setup, this allows a malicious web page to send requests to the local MCP endpoint and gain unrestricted access to the vault via the http-local caller. Version 1.3.1 addresses this issue by enabling proper host, origin, and DNS-rebinding validation.
The npm package @revizahoshii/hoshino-mcp-server contains malicious code that compromises any computer on which it is installed or running. This compromise potentially grants full control to an outside entity, risking exposure of all stored secrets and keys. Immediate removal of the package is advised, but it may not fully eliminate the malicious presence. Secrets and keys should be rotated from a separate, trusted machine to mitigate ongoing risk.
mac-shell-mcp versions before 2.0.0 have an OS command injection vulnerability that allows MCP clients to bypass command whitelisting by injecting shell metacharacters into execute_command arguments. This enables execution of arbitrary commands as the server user without approval.
CVE-2026-108787 is a server-side request forgery (SSRF) vulnerability in houtini-ai voice-analyser-mcp versions 2.1.0 and 2.1.1. The flaw exists in the fetch function of src/utils/crawler.ts within the collect_corpus component, where manipulation of the sitemap_url argument allows SSRF. Remote exploitation is possible, and a public exploit is available. The vendor has not responded to disclosure attempts. The CVSS 4.0 score is 5.3, indicating medium severity.
CVE-2026-108786 is a server-side request forgery (SSRF) vulnerability in the houtini-ai seo-crawler-mcp product, specifically in the run_seo_audit function of src/core/CrawlOrchestrator.ts. It affects versions 2.1.0 through 2.1.5. The vulnerability allows remote attackers to manipulate server requests, potentially causing unauthorized requests to internal or external systems. The vendor has not responded to the disclosure, and no patch or fix is currently available. The CVSS 4.0 score rates this as a medium severity issue.
mcp-go versions up to 1.2.1 have a denial of service vulnerability in the StreamableHTTPServer.ServeHTTP function. Remote unauthenticated attackers can send oversized POST requests that are fully read into memory before validation, allowing them to exhaust server memory and potentially cause the server to crash or become unresponsive.
phpMyFAQ versions 4.1.0 through 4.1.10 contain a missing authorization vulnerability in the MCP server's faq_search tool. This flaw allows MCP clients to bypass user and group permission checks when performing search queries, enabling unauthorized access to restricted FAQs. The vulnerability arises because the Search::searchDatabase() function does not enforce access controls on the returned FAQ content.