MCP Atlassian versions prior to 0.22.0 contain a vulnerability that allows arbitrary server-side file reading via attachment upload. The flaw arises because certain upload tools accept client-controlled file paths without restricting them to an approved server workspace. This enables an attacker with write-tool access to read sensitive files on the MCP server, including environment-held Atlassian credentials or data belonging to other tenants. The issue is fixed in version 0.22.0.
A DNS rebinding time-of-check to time-of-use (TOCTOU) vulnerability in MCP Atlassian (versions >=0.11.9 and <0.22.0) allows an attacker to bypass SSRF protections. The flaw arises because the server resolves attacker-controlled hostnames twice—once during validation and again at connection time without IP pinning—enabling requests to internal or cloud metadata IPs. This vulnerability is fixed in version 0.22.0.
MCP Atlassian versions prior to 0.17.0 contain a Server-Side Request Forgery (SSRF) vulnerability via unvalidated X-Atlassian-Jira-Url and X-Atlassian-Confluence-Url HTTP headers. An unauthenticated attacker able to reach the mcp-atlassian HTTP endpoint can cause the server to make arbitrary outbound HTTP requests. This flaw allows internal network reconnaissance and potential injection of attacker-controlled content into LLM tool results. The vulnerability is fixed in version 0.17.0.
The python-multipart library's parse_form() function did not validate the Content-Length header for negative values, causing it to read the entire request body into memory at once instead of in chunks. This vulnerability only affects code that directly calls parse_form() with attacker-controlled negative Content-Length headers without normalization. Common frameworks like Starlette and FastAPI are not affected. The impact is limited to potential memory exhaustion under concurrent requests in bespoke WSGI or http.server handlers. The issue is fixed in version 0.12.0_2, which rejects negative Content-Length values before reading the stream.
The httplib2 HTTP client library is vulnerable to a decompression bomb denial of service (DoS) attack via unbounded gzip or deflate response handling. A malicious HTTP server can send a small compressed payload that decompresses to a very large size in memory, causing the client process to run out of memory or crash. This affects any application using httplib2.Http().request() with untrusted HTTP endpoints. The vulnerability arises because the library decompresses entire gzip or deflate payloads without size limits or checks. A fix is available in version 0.6.3_4.
MCP Inspector versions 0.14.2 up to but not including 0.16.6 contain a cross-site scripting (XSS) vulnerability. This flaw occurs when the tool renders a redirect URL from a remote MCP server without sufficient validation. If the Inspector connects to an untrusted server, an attacker can craft a redirect that injects malicious script into the Inspector's context. This script can then leverage the built-in proxy to execute arbitrary commands on the developer's machine. Version 0.16.6 includes hardened URL handling and validation that prevents this script execution.
FastMCP's authentication integration with Entra ID allows the MCP server to act both as an OAuth client and an authorization server. Due to Entra ID's lack of support for Dynamic Client Registration, the FastMCP server uses a static app registration for authorization. After a user consents to this static client, Entra ID sets a persistent authorization cookie. An attacker controlling their own MCP client can exploit this by initiating authorization flows that reuse the victim's consented static client ID, potentially leading to account takeover. This vulnerability affects FastMCP versions from 0.11.9 up to but not including 0.11.11 and has a CVSS 4.0 score of 7.3 (high severity). A patch is available in version 0.11.11.
Authlib's OpenID Connect Implicit and Hybrid authorization endpoints contain an unauthenticated open redirect vulnerability. An attacker can craft an authorization request omitting the 'openid' scope, causing the server to redirect users to an attacker-controlled URL. This occurs because the scope validation error redirects using an unvalidated redirect_uri parameter. The flaw affects Authlib versions from 0.11.11 up to but not including 0.21.1_4 and has been fixed in 0.21.1_4. The vulnerability enables phishing and credential harvesting by leveraging the trusted authorization server's domain in the redirect.
Authlib's OAuth 2.0 authorization endpoint contains an open redirect vulnerability triggered by an unsupported response_type parameter combined with an attacker-controlled redirect_uri. This flaw occurs before client validation, allowing unauthenticated attackers to cause the server to redirect to arbitrary URLs. The issue affects versions prior to 0.23.0 and has been confirmed in version 1.6.6 and current HEAD code. A patch is available in version 0.23.0.
FastMCP OpenAPI Provider contains a critical vulnerability in its URL construction logic that allows authenticated attackers to perform server-side request forgery (SSRF) and path traversal. The vulnerability arises because path parameters are substituted into URL templates without URL encoding, enabling attackers to inject traversal sequences like '../' to escape intended API paths. This leads to unauthorized access to internal backend endpoints with the MCP provider's authentication context.
FastMCP OAuthProxy's callback handler does not verify that the browser sending the authorization code has just given consent, enabling a Confused Deputy vulnerability. An attacker can trick a victim's browser, already authorized with GitHub, into redirecting a valid authorization code to a malicious client. This allows the attacker to obtain an access token to the benign MCP server under the victim's GitHub account. The issue affects versions >=0.11.9 and <0.23.0 and is fixed in version 0.23.0.
CVE-2025-69196 is a vulnerability in the FastMCP OAuth Proxy used by MCP servers where the OAuth tokens issued do not properly bind to the specific MCP server resource. Instead, tokens are issued for a base URL, allowing an attacker running a malicious MCP server to steal tokens and access other MCP servers using the same OAuth Proxy. The issue arises because the token does not include the resource parameter, violating the OAuth specification for token audience binding. The vulnerability affects FastMCP versions >=0.11.9 and <0.12.0 and is fixed in version 0.12.0.
A command injection vulnerability exists in FastMCP when installing the claude-code or gemini-cli tools on Windows. Server names containing shell metacharacters such as '&' can cause arbitrary command execution due to how Windows .cmd wrappers are invoked via cmd.exe. This affects FastMCP versions from 0.11.9 up to but not including 0.23.0. The issue does not affect macOS or Linux, nor config-file-based install targets. The vulnerability is patched in version 0.23.0 by validating server names to reject shell metacharacters.
A reflected Cross-Site Scripting (XSS) vulnerability exists in the FastMCP OAuth client's callback page. User-controlled input parameters, such as 'error', are embedded into the callback page without proper escaping or sanitization, allowing arbitrary JavaScript execution in the callback server's origin. This affects FastMCP versions from 0.11.9 up to but not including 0.11.11. The vulnerability has a CVSS 4.0 base score of 5.3 (medium severity).
mcp-server-kubernetes versions 2.9.1 up to but not including 3.5.0 contain an argument injection vulnerability in the port_forward tool. This flaw arises from unsafe string concatenation and naive space splitting of user input when constructing kubectl commands, allowing injection of arbitrary kubectl flags. Exploitation can expose internal Kubernetes services to the network, enable cross-namespace targeting, and facilitate indirect attacks via prompt injection against AI agents connected to the MCP server. The vulnerability is fixed in version 3.5.0.
A command injection vulnerability exists in the exec_in_pod tool of mcp-server-kubernetes versions 2.9.1 through 2.9.7. The tool improperly passes user-supplied commands in string format directly to shell interpretation without validation, allowing execution of arbitrary shell commands. This issue is fixed in version 2.9.8.
Starlette's StaticFiles component on Windows allows server-side request forgery (SSRF) via UNC paths, causing the server to connect to attacker-controlled SMB shares and leak NTLMv2 credentials. This occurs because os.path.realpath resolves UNC paths before access checks, triggering outbound SMB connections. The vulnerability affects default configurations with follow_symlink set to false and impacts applications serving static files on Windows, including those built with FastAPI. Non-Windows systems are not affected. The issue is fixed in version 0.6.3_2.
The kimi-cli FastMCP authentication integration allows a confused deputy attack leading to account takeover. The FastMCP server acts both as an OAuth client to a downstream authorization server (e.g., Entra ID) and as an authorization server to its clients. Due to the use of a static client registration in Entra ID and persistent user consent cookies, an attacker controlling their own MCP client can leverage the victim's prior consent to the static client to bypass re-authorization prompts. This enables the attacker to obtain authorization codes and potentially take over the victim's account. A patch is available for this vulnerability.