CVE-2026-14540 is a Server-Side Request Forgery (SSRF) vulnerability in Google mcp-toolbox versions 0.3.0 through 1.4.0. The vulnerability arises because the HTTP client used by the toolbox does not enforce restrictive redirect policies or validate target IPs, allowing attackers to manipulate redirections. This can cause the toolbox to make unauthorized requests to internal or arbitrary external endpoints. The vulnerability has a high severity score of 8. No official patch or remediation guidance is currently available.
MCP Ruby SDK is the official Ruby SDK for Model Context Protocol servers and clients. Prior to 0.23.0, MCP::Server::Transports::StreamableHTTPTransport in the mcp gem reads and parses an entire JSON-RPC POST body without a size limit, allowing an unauthenticated remote attacker to exhaust process memory. This issue is fixed in version 0.23.0.
MCP Ruby SDK is the official Ruby SDK for Model Context Protocol servers and clients. Prior to 0.23.0, MCP::Server::Transports::StdioTransport and MCP::Client::Stdio in the mcp gem use IO#gets without a byte limit, allowing a peer that sends data without a newline to exhaust process memory. This issue is fixed in version 0.23.0.
CVE-2026-67430 is a medium severity vulnerability in the MCP Ruby SDK prior to version 0.23.0. The issue involves the MCP::Server::Transports::StreamableHTTPTransport component not expiring sessions by default, causing unbounded retention of ServerSession objects. This can lead to memory exhaustion in the process due to repeated initialize requests. The vulnerability is fixed in version 0.23.0.
MCP Ruby SDK is the official Ruby SDK for Model Context Protocol servers and clients. Prior to 0.23.0, MCP::Server::Transports::StreamableHTTPTransport in the mcp gem does not validate the HTTP Host or Origin request headers, which allows a malicious browser page to use DNS rebinding to reach a locally running MCP server and invoke exposed tools. This issue is fixed in version 0.23.0.
In consul-mcp-server, versions 0.1.0 up to 0.1.3 did not properly isolate session state in stateless mode, which may allow one client's Consul authentication token to be used for subsequent requests from other clients. This vulnerability (CVE-2026-16326) is fixed in consul-mcp-server 0.1.4.
The @dynatrace-oss/dynatrace-mcp-server package contains a vulnerability in the create_dynatrace_notebook tool where a required human-approval gate is missing. This omission allows an unauthenticated caller to create persistent tenant-visible notebooks with arbitrary content, including embedded DQL queries that execute under the permissions of any tenant user who opens the notebook. Unlike other write tools in the package, create_dynatrace_notebook does not require operator consent before execution. This vulnerability affects versions prior to 1.8.7.
The terraform-mcp-server before version 1.1.0 is vulnerable to a server-side request forgery issue in the streamable-HTTP transport that may allow an unauthenticated remote client to redirect the server's Terraform API requests, and the server-side authorization token, to an attacker-controlled endpoint. This vulnerability, CVE-2026-14869, is fixed in terraform-mcp-server 1.1.0.
A Server-Side Request Forgery (SSRF) and credential exfiltration vulnerability exists in the cloud-healthcare-fhir-fetch-page tool of googleapis/mcp-toolbox.
The tool takes an unvalidated pageURL parameter from the client and issues an HTTP GET request to it using an authenticated client. The underlying transport automatically attaches an Authorization: Bearer header to every outbound request regardless of the destination host. An attacker can supply an arbitrary external URL to the pageURL parameter (either directly via the tool execution payload or implicitly via data-driven pagination tracking loops), leading Toolbox into sending its OAuth/service-account access token to an attacker-controlled listener. Depending on the configuration, this leaks either the end-user's token or the broader service-account access token (ADC), potentially exposing Protected Health Information (PHI) and secondary Google Cloud Platform services.
Between May and July 2026, security researchers deployed an unauthenticated Model Context Protocol (MCP) honeypot server to observe how threat actors exploit AI agent infrastructure. Of approximately 1,000 sources that reached the decoy, 596 spoke the protocol and 24 proceeded to actively exploit it. These operators executed 628 shell commands, 255 file reads, and 248 secrets-store lookups, with 19 hunting credentials and 4 attempting container escapes. Activity escalated from 39 tool calls in May to 877 by mid-July. Three stolen credentials were subsequently used against a live AWS account, with two cases involving Bedrock model invocation for LLMjacking. The attacks demonstrated automated reconnaissance, credential harvesting, container escape attempts, backdoor account creation, and Kubernetes enumeration, revealing that exposed MCP servers represent a growing attack surface as AI agent infrastructure proliferates.
A vulnerability was detected in nanocoai NanoClaw up to 2.0.64. This impacts the function createChatSdkBridge.setup of the file src/channels/chat-sdk-bridge.ts of the component MCP Server Approval. Performing a manipulation results in improper authorization. The attack needs to be approached locally. The exploit is now public and may be used. The project was informed of the problem early through an issue report but has not responded yet.
In the Linux kernel, the following vulnerability has been resolved:
smb: client: restrict implied bcc[0] exemption to responses without data area
smb2_check_message() has a long-standing quirk that accepts a response
whose calculated length is one byte larger than the bytes actually
received ("server can return one byte more due to implied bcc[0]").
This was introduced to accommodate servers that omit the trailing bcc[0]
overlap byte when no data area is present.
However, the exemption is applied unconditionally, regardless of whether
the command actually carries a data area (has_smb2_data_area[]). When a
response with a data area is subject to the +1 exemption, the reported
data can extend one byte beyond the bytes actually received, yet
smb2_check_message() still accepts it. The subsequent decoder then reads
past the end of the receive buffer. This is reachable during NEGOTIATE
and SESSION_SETUP, before the session is established.
The resulting out-of-bounds reads are visible under KASAN when mounting
against a non-conforming server; both the SPNEGO/negTokenInit and the
NTLMSSP challenge decoders are affected:
BUG: KASAN: slab-out-of-bounds in asn1_ber_decoder+0x16a7/0x1b00
Read of size 1 at addr ffff8880084d67c0 by task mount.cifs/81
CPU: 1 UID: 0 PID: 81 Comm: mount.cifs Not tainted 7.1.0-rc6 #1
Call Trace:
<TASK>
dump_stack_lvl+0x4e/0x70
print_report+0x157/0x4c9
kasan_report+0xce/0x100
asn1_ber_decoder+0x16a7/0x1b00
decode_negTokenInit+0x19/0x30
SMB2_negotiate+0x31d9/0x4c90
cifs_negotiate_protocol+0x1f2/0x3f0
cifs_get_smb_ses+0x93f/0x17e0
cifs_mount_get_session+0x7f/0x3a0
cifs_mount+0xb4/0xcf0
cifs_smb3_do_mount+0x23a/0x1500
smb3_get_tree+0x3b0/0x630
vfs_get_tree+0x82/0x2d0
fc_mount+0x10/0x1b0
path_mount+0x50d/0x1de0
__x64_sys_mount+0x20b/0x270
do_syscall_64+0xee/0x590
entry_SYSCALL_64_after_hwframe+0x77/0x7f
</TASK>
Allocated by task 85:
kmem_cache_alloc_noprof+0x106/0x380
mempool_alloc_noprof+0x116/0x1e0
cifs_small_buf_get+0x31/0x80
allocate_buffers+0x10d/0x2b0
cifs_demultiplex_thread+0x1d5/0x1d50
kthread+0x2c6/0x390
ret_from_fork+0x36e/0x5a0
ret_from_fork_asm+0x1a/0x30
The buggy address is located 0 bytes to the right of
allocated 448-byte region [ffff8880084d6600, ffff8880084d67c0)
which belongs to the cache cifs_small_rq of size 448
BUG: KASAN: slab-out-of-bounds in kmemdup_noprof+0x36/0x50
Read of size 329 at addr ffff88800726c678 by task mount.cifs/89
CPU: 0 UID: 0 PID: 89 Comm: mount.cifs Tainted: G B 7.1.0-rc6 #1
Call Trace:
<TASK>
dump_stack_lvl+0x4e/0x70
print_report+0x157/0x4c9
kasan_report+0xce/0x100
kasan_check_range+0x10f/0x1e0
__asan_memcpy+0x23/0x60
kmemdup_noprof+0x36/0x50
decode_ntlmssp_challenge+0x457/0x680
SMB2_sess_auth_rawntlmssp_negotiate+0x6f0/0xcb0
SMB2_sess_setup+0x219/0x4f0
cifs_setup_session+0x248/0xaf0
cifs_get_smb_ses+0xf79/0x17e0
cifs_mount_get_session+0x7f/0x3a0
cifs_mount+0xb4/0xcf0
cifs_smb3_do_mount+0x23a/0x1500
smb3_get_tree+0x3b0/0x630
vfs_get_tree+0x82/0x2d0
fc_mount+0x10/0x1b0
path_mount+0x50d/0x1de0
__x64_sys_mount+0x20b/0x270
do_syscall_64+0xee/0x590
entry_SYSCALL_64_after_hwframe+0x77/0x7f
</TASK>
Allocated by task 93:
kmem_cache_alloc_noprof+0x106/0x380
mempool_alloc_noprof+0x116/0x1e0
cifs_small_buf_get+0x31/0x80
allocate_buffers+0x10d/0x2b0
cifs_demultiplex_thread+0x1d5/0x1d50
kthread+0x2c6/0x390
ret_from_fork+0x36e/0x5a0
ret_from_fork_asm+0x1a/0x30
The buggy address is located 120 bytes inside of
allocated 448-byte region [ffff88800726c600, ffff88800726c7c0)
which belongs to the cache cifs_small_rq of size 448
Restrict the +1 exemption to responses that have no data area, so that
it still covers the bcc[0] omission it was meant for. When a data area
is present, the +1 discrepancy instead means the reported data length
overruns the
---truncated---
SiYuan before v3.7.2 contains a missing authorization vulnerability in the POST /mcp kernel endpoint, which is gated only by a general auth check (model.CheckAuth) with no admin-role or read-only enforcement. This exposes 31 MCP tools, including a file tool with list/read/write/delete/rename/copy actions across the entire workspace. When the Publish server is enabled in anonymous mode (Conf.Publish.Enable=true and Conf.Publish.Auth.Enable=false), the Publish reverse proxy attaches an anonymous RoleReader JWT to proxied requests, allowing a remote unauthenticated attacker to reach /mcp. The attacker can read conf/conf.json to extract accessAuthCode, api.token, and cookieKey in plaintext, write arbitrary files in the workspace, and plant a plugin into data/plugins/ that executes with nodeIntegration:true and no contextIsolation on the next desktop launch, leading to administrator takeover.
In the Linux kernel, the following vulnerability has been resolved: fuse-uring: make a fuse_req on SQE commit only findable after memcpy Bad userspace might try to trick us and send commit SQEs request unique / commit-id of requests that are not even send to fuse-server (io_uring_cmd_done() not called) yet. fuse_uring_commit_fetch() ends the fuse request when the ring entry has a wrong state, but that could have caused a use-after-free with the memcpy operations in fuse_uring_send_in_task(). In order to avoid such races the call of fuse_uring_add_to_pq() is moved after the copy operations and just before completing the io-uring request - malicious userspace cannot find the request anymore until all prepration work in fuse-client/kernel is completed. This also moves fuse_uring_add_to_pq() a bit up in the code to avoid a forward declaration. Also not with a preparation commit, to make it easier to back port to older kernels.
The OpenAPI adapter in @frontmcp/adapters versions prior to 1.5.6 contains a Server-Side Request Forgery (SSRF) vulnerability in its spec-change poller. While the initial spec load is protected by an SSRF guard, the recurring poller fetches the spec URL using an unguarded global fetch, bypassing protections such as allow-list/block-list enforcement, internal IP blocking, DNS resolution pinning, and redirect validation. This allows an attacker who can influence the spec URL to cause the server to make unauthorized GET requests to internal or otherwise protected network resources. The vulnerability primarily impacts confidentiality by enabling reading of internal endpoints, including cloud metadata services. The issue is fixed in version 1.5.6 by routing the poller's fetch through the same SSRF guard as the initial load.
Jan through 0.8.4, fixed in commit 3e1c1e7, contains a CORS misconfiguration vulnerability in its local API server that allows network-adjacent attackers to bypass trusted host restrictions by exploiting the server's replacement of user-configured trusted hosts with a wildcard that reflects arbitrary origins with credentials. Attackers on the local network or using DNS rebinding can reach the unauthenticated OpenAI-compatible API to perform inference, enumerate models, invoke MCP tools, and read cross-origin responses.
APIFold reads an OpenAPI 3.x or Swagger 2.x specification and generates a live, production-ready MCP server endpoint. Prior to commit 7f19b52280f414f57af2b79a95333d1c8fbeece5, the `/webhooks/:serverSlug/:eventName` endpoint accepts arbitrary unauthenticated JSON and stores it in Redis and the `webhook_events` PostgreSQL table without any signature check or authentication requirement. The root cause is that `createWebhookRouter` is called at `server.ts:188` without a `validators` map, so `receivers.ts:80`'s optional-chaining guard evaluates to `undefined` and the signature-validation block (`receiver.ts:81–95`) is unconditionally skipped. Any unauthenticated network client that knows a valid server slug can inject arbitrary payloads, which are subsequently served as trusted resource state to legitimate MCP clients. Commit 7f19b52280f414f57af2b79a95333d1c8fbeece5 patches the issue.
Office-Word-MCP-Server through 1.1.11 contains a path traversal vulnerability in its document tools that allows attackers who can influence the filename argument to read arbitrary .docx files or create and overwrite .docx files outside the intended working directory. Attackers can supply absolute paths or ../ traversal sequences directly to document open and save operations, bypassing the check_file_writeable and ensure_docx_extension helpers which perform no base-directory confinement or realpath validation.