CVE Database

OpenRemote versions before 1.28.0 contain a cross-realm information disclosure vulnerability in the Notification REST API that allows per-realm tenant administrators to read all tenants' sent notifications including message bodies. Attackers with read:admin credentials in one realm can submit a zero-parameter GET request to the notification endpoint to retrieve sensitive notification metadata and message content from all realms.

AVideo before 24.0 contains a server-side request forgery vulnerability in the isSSRFSafeURL function that fails to extract embedded IPv4 addresses from NAT64, 6to4, and Teredo IPv6 transition address formats. Unauthenticated attackers can bypass SSRF protections via the LiveLinks proxy endpoint to reach internal services and cloud metadata endpoints by encoding private IPv4 targets in transition address formats.

The OpenFaaS gateway registers GET /system/telemetry in gateway/main.go and, when basic_auth is enabled, wraps each administrative /system/* handler in auth.DecorateWithBasicAuth. TelemetryHandler was left out of that wrap block from 0.27.11, which introduced the route, until 0.27.14, which added it. On an affected gateway the route therefore reaches the forwarding proxy with no credential check and returns whatever the configured provider serves for /system/telemetry, so any client that can reach the gateway port can read the provider's telemetry regardless of the basic_auth setting. The exposed content depends on the provider, and covers resource and invocation metrics for faasd and pod or cluster state for faas-netes.

Baserow dispatches an Application Builder data source without acting on the result of its permission check. The dispatch and record-name views in backend/src/baserow/contrib/builder/api/data_sources/views.py are declared with a permission class that admits any caller, so a request carrying no credential reaches the handler. DataSourceService.dispatch_data_sources in backend/src/baserow/contrib/builder/data_sources/service.py then calls check_multiple_permissions without asking it to raise, and neither stores nor examines the mapping of denials it returns, so a denied check leaves execution to continue and the data source is dispatched whatever the caller's identity. The dispatch runs with the integration's own credentials, so an unauthenticated request naming a data source receives the rows and fields that source reads. Identifiers are small integers and can be enumerated. Version 2.3.1 passes raise_exception to the same call.

darknet subscripts its layer array with an index taken from a configuration file without checking it against the array's length. The array is allocated in src-lib/darknet_network.cpp as xcalloc(net.n, sizeof(Darknet::Layer)), sized to exactly the number of layer sections the file declares. The shortcut, scale_channels and sam sections supply that index through their from field and the route section through its layers field, and parse_shortcut_section in src-lib/darknet_cfg.cpp reads net.layers[index].outputs with no bounds check, which reads past the allocation. The dispatch loop in create_network then reuses the same index to assign net.layers[l.index].use_bin_output and net.layers[l.index].keep_delta_gpu, writing past the allocation at an offset the file controls, with a fixed one-byte value. Parsing a crafted configuration file is sufficient: the parse runs before any weights file is opened and needs no non-default option, so the result is a reliable crash and a write whose location, though not its value, is chosen by whoever supplied the file.

The Dash MCP server bound its listener to the loopback address but never checked the host a request named. src/mcp_server_dash.py constructed the server for its network mode with the interface restricted to loopback and no transport-security settings, so a name that had been pointed at the loopback address still reached the listener while carrying the attacker's host name. A page in a visitor's browser could therefore drive the local server and invoke its company-search and file-detail tools under the Dropbox credential the server holds. Only the network mode was reachable this way; the standard input mode was not. The fix supplies transport-security settings that enable host checking and allow only the loopback name and port, rejecting other hosts before a tool runs. The repository publishes no versions, so the affected boundary is the commit preceding the fix.

The configure command accepted any endpoint URL and stored it beside the user's access token. ConfigureCommand.execute in src/cli.ts persisted the value given to its endpoint option into the user profile without passing it through createSafeUrl in src/config.ts, the helper that already restricted the environment-variable form of the same setting to the vendor's own hosts over HTTPS. Because the connect path in src/mcp.ts attaches the stored token as a bearer credential on every request to the configured endpoint, a user who was persuaded to run configure with an endpoint of the attacker's choosing sent their personal access token to that destination on each subsequent invocation. Version 0.2.5 applies the same helper to the option.

tiger-gh-mcp-server started its MCP HTTP transport without enabling the host allow-list the underlying SDK provides. src/httpServer.ts called the shared httpServerFactory helper and never set the DNS-rebinding-protection option, so the transport accepted a request whatever host it named, making the locally reachable GitHub MCP endpoint drivable from a page in a visitor's browser that pointed a name it controlled at the bound address. The fix passes the option explicitly alongside a dependency update; the update alone would not have closed it. The repository has published no release that brackets the fix, so the affected boundary is the commit preceding it.

tiger-slack started its MCP HTTP transport without enabling the host allow-list the underlying SDK provides. mcp/src/httpServer.ts called the shared httpServerFactory helper and never set the DNS-rebinding-protection option, so the transport accepted a request whatever host it named, and a page in a browser could point a name it controlled at the address the server was bound to and drive the locally reachable Slack MCP server through the visitor's browser. The fix passes the option explicitly alongside a dependency update; the update alone would not have closed it. The repository publishes no release that brackets the fix, so the affected boundary is the commit preceding it.

The Telnyx MCP server exposed its HTTP transport on every interface and did not require a caller credential. packages/mcp-server/src/http.ts served MCP on the root path with a listener bound to all interfaces and parsed the caller's authentication headers in a mode that did not fail when they were absent, so a request without any credential completed initialisation and dispatched tools. Dispatch forwarded the server's own stored credentials, the Telnyx API key and client secret together with the code-execution key, to the upstream endpoint, so an unauthenticated caller able to reach the port acted with them. The current code defaults the host to loopback, requires a server API key, and enforces it in middleware.

The execute_ruby tool is documented as a read-only Ruby sandbox and is enforced by a pattern denylist together with replacements for the process-spawning methods on Kernel. The pseudo-terminal library's spawn entry points are neither in the denylist nor replaced, so a normal tool call could reach them and start a shell, executing commands as the account running the server and outside the guarded methods. The denylist was introduced with the tool in 1.4.0 and never covered those entry points through 1.6.0. Version 1.6.1 restricts the requires the sandbox permits to a data-only list and blocks dynamic dispatch to execution entry points; 2.0.0 removes the tool.

ToolUniverse ran caller-supplied Python inside a sandbox that could be escaped, on a server that required no authentication. The executor behind the python_code_executor tool, in python_executor_tool.py, inspected the submitted source for a denied list of attribute names and calls but left the attribute-lookup builtins available and did not stop a dunder attribute reached through a string lookup or through a module already permitted, so a caller could walk from a literal's class to its base and enumerate subclasses to obtain a reference to the process and subprocess modules. A per-call argument also let the caller widen the import allow-list before the inspection ran. The HTTP and MCP servers in http_api_server.py and smcp_server.py bound to every interface with debugging enabled and no authentication, so any caller able to reach the port executed code as the server process. Version 1.3.0 adds bearer-token authentication, defaults the bind address to loopback, and hardens the attribute checks.

pg-aiguide started its MCP HTTP transport without enabling the host allow-list the underlying SDK provides. src/httpServer.ts called the shared httpServerFactory helper and never set the DNS-rebinding-protection option, so the transport accepted a request whatever host it named. A page in a browser could therefore point a name it controlled at the address the server was bound to and drive the locally reachable MCP server through the visitor's browser. The protection was already available in the packaged transport and simply not turned on, so updating the dependency alone would not have closed it. Version 0.5.1 passes the option explicitly.

The mcp-router CLI served its MCP aggregator on every interface and enforced authentication only when the operator asked for it. The serve command in apps/cli/src/commands/serve.ts defaulted its host to the all-interfaces address on a fixed port, and required a token only when the corresponding flag was supplied, so a default invocation exposed the aggregator, and every MCP server it fronted, to anyone able to reach the port. Release 0.6.3 defaults the host to the loopback address and refuses to start without a token whenever the host it is given is not a loopback address; no earlier release carries either check.

The get-html-skeleton tool fetched a URL the caller supplied after checking only its syntax. The handler in src/tools/common/get_html_skeleton.ts validated the url argument with isValidHttpUrl from src/utils/generic.ts, which confirmed the string began with an http or https scheme and parsed as a URL and inspected neither the host name nor the address it resolves to. Loopback, link-local and private ranges therefore passed, including the address cloud providers use to serve instance metadata. The unchecked URL was handed to the web-browser actor and the fetched document was returned in the tool response, so any caller of the MCP server could make it request an endpoint reachable only from the host and read the result, including instance credentials. Version 0.9.12 removes the tool.

mcp-go accepted requests on its HTTP transports without checking the Host header. StreamableHTTPServer.ServeHTTP in server/streamable_http.go and SSEServer.ServeHTTP in server/sse.go served any request arriving over a loopback connection regardless of the host it named, and the SSE transport's cross-origin default allowed any origin. A page in a browser could therefore point a name it controlled at the loopback address and reach a server listening there, invoking tools and reading resources that the server exposed on the assumption that only local software could connect. No release before 0.56.0 validated the header on either transport; 0.56.0 adds server/http_localhost.go, which rejects a loopback-bound request carrying a host that is not a loopback name, and wires it into both transports.

The proxy middleware in mcp-use's inspector forwards requests to a destination the caller names. mountMcpProxy in libraries/typescript/packages/inspector/src/server/proxy/mcp-proxy.ts read the target from the X-Target-URL header or the __mcp_target parameter and proxied to it without inspecting the host, so loopback, link-local and private addresses were all accepted, as were names that resolve to them, and the validation was not reapplied to a redirect the destination returned. A caller could therefore make the server issue requests to addresses reachable only from the host it runs on and read the responses. The current code calls isSafeProxyTarget, which checks the resolved address against private, loopback and link-local ranges before proxying and bounds the number of redirects followed.

An issue was discovered in the resolv gem before 0.7.2 for Ruby. Resolv::DNS::MessageEncoder wrote a DNS label's length into a single octet without checking its range. A label longer than 255 octets had its length stored modulo 256 but the label data was written unchanged, and thus the bytes on the wire described a different name than the one the application asked to encode. RFC 1035 section 2.3.4 limits a label to 63 octets, and the two high bits of the length octet are reserved for compression pointers. put_string packed the length with put_pack("C", d.length) and put_label used it for labels, and thus any value from 0 to 255 could end up as a label length octet, including the reserved 0x40-0xBF range and the 0xC0-0xFF pointer range. Resolv::DNS::Name.create did not check per-label or total name length either, and thus an attacker-controlled hostname reached the encoder unchanged. An application that resolves an attacker-controlled hostname sends a query whose wire bytes name a domain the attacker chose. A hostname suffix that the application validates against an allowlist becomes padding that never appears on the wire, and thus allowlist and egress checks can be bypassed. The recursive resolver caches the response under the attacker's name, and DNS logs record that name rather than the one the application asked for. A label length whose low octet lands in the 0xC0-0xFF range produces a length octet that conforming parsers read as the start of a compression pointer, with the following attacker-controlled byte as the offset.

An issue was discovered in the resolv gem before 0.7.2 for Ruby. Resolv::DNS::Resource.get_class, Resolv::DNS::Resource::Generic.create, and Resolv::DNS::SvcParam::Generic.create generate a new class for each unknown DNS resource record (type, class) pair, or each unknown SvcParamKey, encountered while decoding a response. Each generated class was permanently registered both as a constant on Resource (or SvcParam::Generic) and as an entry in a class-lookup hash (ClassHash), and thus the class remained reachable through that constant after the response was discarded. Type and class are each 16-bit values, and thus an attacker controlling DNS responses (a spoofed response, or a malicious or hijacked upstream DNS server) has roughly 2^32 distinct (type, class) pairs to choose from. A single response of a few hundred kilobytes carrying tens of thousands of distinct unknown types permanently grows process memory by tens of megabytes; repeated responses accumulate without bound and are never reclaimed by garbage collection, because the constant keeps each class alive. Any code path that calls Resolv::DNS::Message.decode on attacker-influenced DNS responses is affected. resolv is a default gem, and thus this is reachable from a plain Ruby installation without any additional dependency.

FrontAccounting through 2.4.20 stores and verifies user passwords as unsalted MD5 digests. admin/users.php passes md5($_POST['password']) to add_user() and update_user_password(), admin/change_current_user_password.php does the same when a user changes their own password, the forgotten-password path in includes/current_user.inc hashes the newly generated password the same way, and authentication calls get_user_auth($loginname, md5($password)). The codebase applies no per-password salt and contains no call to password_hash(), password_verify() or any other adaptive hash, so identical passwords yield identical digests and an attacker who obtains the user table can recover plaintext passwords with precomputed lookup tables or high-rate GPU cracking.

FrontAccounting through 2.4.20 generates a CSRF token in end_form() in includes/ui/ui_controls.inc and embeds it as the _token hidden field in every form it renders, but only admin/users.php and admin/change_current_user_password.php call check_csrf_token() to validate it. No financial transaction handler validates the token, including gl/gl_journal.php, gl/gl_bank.php, purchasing/supplier_invoice.php, sales/customer_invoice.php, sales/customer_payments.php and admin/company_preferences.php, so those endpoints act on POST data with no origin check. An attacker who gets an authenticated user to load a page under attacker control can auto-submit a cross-origin form to any of them and have the forged journal entry, invoice, customer payment, bank transaction or company configuration change recorded under the victim's session.

The updateWorkspace handler in mods/identity/src/workspaces/createUpdateWorkspace.ts in Fonoster through 0.22.7 invokes the gRPC callback with PERMISSION_DENIED when createIsWorkspaceMember reports that the caller is not a member of the target workspace, but it does not return. Execution continues into prisma.workspace.update, which is scoped by the workspace reference alone, so the rename commits before the second callback is issued and the caller receives a permission error for a write that already succeeded. The gRPC interceptor in mods/common/src/identity/createAuthInterceptor.ts binds the workspace accessKeyId to the caller's token only for paths in workspaceResourceAccess or workspaceResourceOwnerOrAdminAccess, and Identity/UpdateWorkspace is listed in fullIdentityAccess, which the base USER role holds. Any authenticated user can therefore rename an arbitrary workspace in the deployment.

APITable through 1.13.0-beta.1 annotates both getUserHistories and closePausedUserAccount in InternalUserController with requiredLogin = false. ResourceInterceptor honours that annotation by returning before any session or API key is validated, and the nginx gateway shipped with the product proxies every /api request to the backend server, so both endpoints are reachable by any unauthenticated client that can reach the gateway. An attacker can POST to /api/v1/internal/getUserHistories to enumerate the accounts sitting in the 30-day cooling-off period that follows a deletion request, then POST to /api/v1/internal/users/{userId}/close for each one. The closure path clears the account's email address, phone number and nickname, cancels its space subscriptions, removes its space memberships and deletes its OAuth bindings, so the cooling-off window that exists to let a user reverse a deletion request is bypassed and the account cannot be recovered.

APITable through 1.13.0-beta.1 annotates the create handler of InternalNotifyController with requiredLogin = false. ResourceInterceptor honours that annotation by returning before any session or API key is validated, and the nginx gateway shipped with the product proxies every /api request to the backend server, so POST /api/v1/internal/notification/create is reachable by any unauthenticated client that can reach the gateway. An attacker can persist arbitrary notifications in apitable_player_notification against any user ID they name, and because fromUserId can be set to 0 the message is stored with the system sender and renders in the victim's inbox as a legitimate system notification. The body extras content is stored verbatim and forwarded to the frontend.

The Twig sandbox mechanism in Craft CMS is configured to allow dangerous functionality from the Yii framework, leading to authenticated RCE similar to previously disclosed vulnerabilities.

Reflected XSS in Netron versions <=9.1.2 on desktop application through unsanitized node names allows an attacker to hide certain nodes, perform port scanning or abuse a Chrome n-day to achieve Remote Code Execution.

Reflected XSS in Netron versions <=9.1.2 on desktop application through unsanitized node names allows an attacker to hide certain nodes, perform port scanning or abuse a Chrome n-day to achieve Remote Code Execution.

Reflected XSS in Netron versions <=9.1.2 on desktop application through unsanitized node names allows an attacker to hide certain nodes, perform port scanning or abuse a Chrome n-day to achieve Remote Code Execution.

In Eclipse SW360 versions 19.0.0, 19.1.0, 19.2.0, 20.0.0, 20.1.0, if the system is configured to use file system storage with config key enable.attachment.store.to.file.system, the attacker can manipulate the filename upon upload and can essentially cause arbitrary file path traversal. The immediate workaround is to disable enable.attachment.store.to.file.system or update to fixed versions.

DJI drones contain an FTP service that uses hardcoded credentials shared across affected models and permits authenticated users to upload files without limits on file size, file count, or total storage consumed in **/blackbox/upgrade/**, as well as overwrite existing files in that directory. An attacker with access to the drone's internal network or USB RNDIS interface can exhaust the available storage, preventing the aircraft from writing flight records, logs, and telemetry and potentially preventing subsequent firmware updates. Uploaded files persist across reboot and factory reset. Affected models are DJI Neo until 01.00.0400, DJI Neo 2 until 01.00.0500, DJI Flip until 01.00.1200, DJI Air 3 until 01.00.1600, DJI Air 3S until 01.00.1400, DJI Avata 2 until 01.00.0400, DJI Avata 360 until 01.00.0300, DJI Mavic 3 until 01.00.1400, DJI Mavic 3 Classic until 01.00.0800, DJI Mavic 3 Pro until 01.01.0700, DJI Mavic 4 Pro until 01.00.0500, DJI Mini 2 until 01.07.0200, DJI Mini 3 until 01.00.0500, DJI Mini 3 Pro until 01.00.0900, DJI Mini 4 Pro until 01.00.1100, and DJI Mini 5 Pro until 01.00.0600. Remediation requires a firmware update from the vendor.