MCPVault is a lightweight Model Context Protocol server for safe access to files in an Obsidian vault. Prior to 0.11.4, PathFilter in src/pathfilter.ts compiles restricted-directory patterns case-sensitively and compares paths without canonicalizing filesystem-equivalent segment names. On case-insensitive macOS and Windows filesystems, case variants of .git, .obsidian, or node_modules pass both isAllowed() and isAllowedForListing() even though the operating system opens the restricted directory, and Windows trailing dots or spaces provide the same bypass. An attacker who influences a path selected by an AI agent can use the bypass in read, write, move, search, or listing operations to expose or modify sensitive repository and Obsidian metadata. Vault-root .. containment is not affected. This issue is fixed in version 0.11.4.
mcp-searxng is a Model Context Protocol server that gives AI assistants web search and URL-reading capabilities through SearXNG. Prior to 1.7.1, web_url_read in src/index.ts passes a caller-supplied URL to readUrlContent() in src/url-reader.ts, where checkContentLength() treats a missing Content-Length header as an inconclusive preflight and the normal and error paths then consume the complete body with response.text(). A server that omits Content-Length can therefore bypass URL_READ_MAX_CONTENT_LENGTH_BYTES and force unbounded memory use. The resulting string is also processed by NodeHtmlMarkdown.translate(), increasing CPU consumption and allowing an unauthenticated HTTP client to cause denial of service. This issue is fixed in version 1.7.1.
mcp-searxng is a Model Context Protocol server that gives AI assistants web search and URL-reading capabilities through SearXNG. Prior to 1.7.1, web_url_read receives its caller-controlled URL through src/index.ts and validates only the literal hostname in assertUrlAllowed() within src/url-reader.ts before undiciFetch() performs operating-system DNS resolution. A public-looking attacker-controlled hostname that resolves to a private, loopback, link-local, or cloud-metadata address therefore passes the lexical check and causes the MCP server to connect to the internal destination. In the default HTTP configuration, an unauthenticated network client can use this path to read internal services, expose credentials or service tokens, and enumerate reachable internal hosts; in STDIO deployments, prompt-influenced tool selection can provide the malicious URL. Direct private IP literals are blocked, and MCP_HTTP_ALLOW_PRIVATE_URLS remains an explicit opt-out. This issue is fixed in version 1.7.1.
mcp-searxng is a Model Context Protocol server that gives AI assistants web search and URL-reading capabilities through SearXNG. Prior to 1.2.0, web_url_read passes a caller-supplied URL to the server-side fetch path while assertUrlAllowed() in src/url-reader.ts runs only when MCP_HTTP_HARDEN is enabled, even though MCP_HTTP_HARDEN is disabled by default in src/http-security.ts. In the default configuration, an attacker who influences the URL selected by a user or AI agent can make the server fetch loopback, private-network, or cloud metadata endpoint resources and return their contents into the model context. file:// URLs remain rejected, and the separate DNS-resolution and redirect-validation bypasses are outside this record. This issue is fixed in version 1.2.0.
mcp-searxng is a Model Context Protocol server that gives AI assistants web search and URL-reading capabilities through SearXNG. Prior to 1.2.0, the web_url_read URL policy in src/url-reader.ts can be bypassed while MCP_HTTP_HARDEN is enabled and MCP_HTTP_ALLOW_PRIVATE_URLS is not enabled because redirect targets are not revalidated, 0.0.0.0 is not classified as an internal address, and IPv4-mapped IPv6 literals canonicalized to hexadecimal form are not recognized. These inputs allow an attacker-influenced tool call to make the MCP server fetch loopback or internal HTTP resources and return content from local services, private APIs, service-mesh endpoints, or cloud metadata endpoints. The separate hostname-to-private-address case addressed by the earlier partial fix is not part of these residual bypasses. This issue is fixed in version 1.2.0.
CVE-2026-54549 is a server-side request forgery (SSRF) vulnerability in the meta-ads-mcp product by pipeboard-co. The flaw exists in versions prior to 1.0.115 in the upload_ad_image tool, which improperly handles attacker-controlled URLs without validating scheme, host, or resolved IP address. This allows an attacker to make the server perform unauthorized HTTP requests to internal or private network resources. The vulnerability is fixed in version 1.0.115.
Meta Ads MCP is a Model Context Protocol (MCP) server that lets AI assistants run Meta Ads. Prior to version 1.0.115, AuthInjectionMiddleware in meta_ads_mcp/core/http_auth_integration.py rejects HTTP MCP requests only when both auth_token and pipeboard_token are absent, while extract_token_from_headers() does not recognize X-Pipeboard-Token as a primary credential. A network caller using the streamable-http transport can therefore send any X-Pipeboard-Token value, pass the guard without establishing authentication context, and cause get_auth_token() to fall back to the server operator's META_ACCESS_TOKEN. Subsequent MCP tools execute with the operator's Meta credentials and can read or modify the operator's Meta Ads data. Deployments using the default stdio transport or without META_ACCESS_TOKEN are not affected. This issue is fixed in version 1.0.115.
Lokka is a Model Context Protocol server for Microsoft 365, including Microsoft Graph and other services. Prior to 2.1.2, the Lokka-Microsoft tool in src/mcp/src/main.ts uses direct URL string concatenation to append the user-controlled path value to the management.azure.com base URL. A specially crafted path can alter URL authority parsing and cause an Azure Resource Manager bearer token to be sent to an unintended host. This issue is fixed in version 2.1.2.
MCP Context Forge is an AI gateway, registry, and proxy for MCP, A2A, REST, and gRPC APIs. Prior to 1.0.2, the python_sandbox_server in mcp-servers/python/python_sandbox_server/src/python_sandbox_server/server_fastmcp.py exposes raw getattr through safe_builtins, omits a required _getattr_ guard, and relies on validate_code checks for literal dangerous dunder strings. An attacker can construct dunder names at runtime, traverse the Python class hierarchy, reach subprocess.Popen, and execute OS commands with the server process privileges through the execute_code MCP tool. The HTTP/SSE transport can expose this tool without authentication, while stdio-only deployments have reduced network reachability. The issue affects the python_sandbox_server subproject and does not directly affect the core Context Forge gateway or proxy components. This issue is fixed in version 1.0.2.
MCP Memory Keeper is an MCP server for persistent context management in AI coding assistants. Prior to 0.13.0, context_import in src/index.ts passes the caller-controlled filePath directly to fs.readFileSync without restricting the path to an export directory. An MCP client, including an LLM agent induced to call the tool, can use ../ traversal or an absolute path to target any file readable by the server process. A valid JSON file is parsed and imported into the caller's session, allowing its full contents to be retrieved through context_get or context_export, while JSON.parse errors for non-JSON files can return leading file bytes in a SyntaxError message. The two disclosure modes can expose other exported sessions, JSON credentials or service-account files, environment files, and portions of SSH keys or other local files. This issue is fixed in version 0.13.0.
CVE-2026-58196 is a Server-Side Request Forgery (SSRF) vulnerability in stacklok's toolhive utility versions prior to 0.31.0. The flaw arises from the handling of authentication discovery URLs, where the server-side HTTP clients trust remote-server-controlled destinations and follow redirects without proper restrictions. This allows a malicious or compromised MCP server to induce the toolhive client to make unauthorized requests to internal-only HTTP services, including link-local and RFC1918 addresses. The vulnerability can be exploited to access internal metadata services such as AWS IMDSv1, potentially leaking sensitive credentials. The issue is fixed in version 0.31.0.
atomic-agents-stack before 1.1.0 accepts cleartext HTTP schemes in the HTTP MCP server-registry backend factory, allowing network man-in-the-middle attackers to rewrite catalog responses. Attackers can inject arbitrary command and argument values that are spawned as local subprocesses by MCPClientPool to achieve code execution on the agent host.
Flowise before 3.1.4 contains a remote code execution vulnerability in the Custom MCP node that allows authenticated attackers to execute arbitrary code by supplying npx package names in the mcpServerConfig parameter. Attackers can invoke npx with attacker-controlled npm packages to execute code on the Flowise server.
CVE-2026-59973 is a Server-Side Request Forgery (SSRF) vulnerability in the agentfront frontmcp framework affecting versions prior to 1.5.0 and 2.5.0. It allows an authenticated user with the ability to import or configure OpenAPI specifications to cause backend-origin requests to internal services by exploiting insufficient validation of external $ref URLs. This can expose internal administrative APIs and private network endpoints. The vulnerability is fixed in frontmcp and related packages version 1.5.0 and mcp-from-openapi 2.5.0.
Contentful MCP Server is a Model Context Protocol server for the Contentful Management API. Prior to @contentful/mcp-server 1.7.19 and @contentful/mcp-tools 0.4.5, export_space and import_space in packages/mcp-tools/src/tools/jobs/space-to-space-migration/exportSpace.ts and packages/mcp-tools/src/tools/jobs/space-to-space-migration/importSpace.ts expose host, proxy, rawProxy, and insecure network options to LLM-controlled tool arguments and combine those options with the server's CONTENTFUL_MANAGEMENT_TOKEN. After space_to_space_migration_handler enables the migration tools, a direct MCP call or prompt injection through attacker-controlled Contentful content can redirect Contentful Management API requests and their Authorization header to an attacker-controlled host or proxy. The regular tools that use createToolClient are unaffected because those tools pin the host from server configuration. Exposure of the personal access token permits persistent out-of-band access to every Contentful space within the token's scope. This issue is fixed in @contentful/mcp-server 1.7.19 and @contentful/mcp-tools 0.4.5.
CVE-2026-59971 is a critical vulnerability in designcomputer's mysql_mcp_server prior to version 0.4.2. The issue arises when MCP_TRANSPORT is set to sse, causing the server to construct an insecure transport without authentication or DNS rebinding protection. This allows unauthenticated network attackers to execute arbitrary SQL queries, potentially disclosing or modifying database contents. If the MySQL account has FILE privileges, attackers can read or write server files and possibly execute code. The vulnerability is fixed in version 0.4.2.
PraisonAI is a multi-agent teams system. From 1.5.0 until 1.7.2, MCPServer.startHttp() in src/praisonai-ts/src/mcp/server.ts binds without a host restriction and forwards every HTTP POST request to handleRequest() without authentication or authorization. Any network client that can reach the port can call tools/list, tools/call, resources/read, or prompts/get, causing registered handlers to run with server-side credentials and process privileges or disclose registered data. An initial remediation was released in version 1.7.2.
IBM Langflow OSS versions 1.0.0 through 1.11.5 contain a vulnerability in the MCP Tools component where improper cache key isolation allows an authenticated attacker to access another user's MCP server context. This issue could lead to limited confidentiality and integrity impacts but does not affect availability.
MCP Atlassian is a Model Context Protocol (MCP) server for Atlassian products (Confluence and Jira). From 0.17.0 until 0.22.0, validate_url_for_ssrf resolves the attacker-controlled X-Atlassian-Jira-Url and X-Atlassian-Confluence-Url header host once at middleware time, but the outbound request is built with the raw hostname and resolves it again at connection time with no IP pinning. An attacker-controlled DNS-rebinding name can return a public IP during validation and 169.254.169.254 or another internal IP during connection, enabling unauthenticated server-side requests to cloud metadata or internal services. The flaw spans src/mcp_atlassian/utils/urls.py, src/mcp_atlassian/servers/main.py, and src/mcp_atlassian/servers/dependencies.py; validate_url_for_ssrf returns only a verdict rather than a pinned IP, UserTokenMiddleware processes the attacker-controlled headers before fetcher creation, and the Jira and Confluence fetchers use the raw hostname. This issue is fixed in version 0.22.0.
MCP Atlassian is a Model Context Protocol (MCP) server for Atlassian products (Confluence and Jira). Prior to 0.22.0, the confluence_upload_attachment and confluence_upload_attachments tools pass a client-controlled file_path through src/mcp_atlassian/confluence/attachments.py upload_attachment, and the jira_update_issue attachments parameter reaches src/mcp_atlassian/jira/attachments.py upload_attachment, without confining either path to an approved server workspace. In a remote HTTP, SSE, or multi-user deployment, absolute or traversing paths are resolved on the MCP server and uploaded to Atlassian, allowing a client with write-tool access to disclose server files, environment-held Atlassian credentials, or another tenant's data. A local single-user stdio deployment does not cross this trust boundary because the server runs in the caller's environment. This issue is fixed in version 0.22.0.