MaxKB is an open-source AI assistant for enterprise. From version 2.0.0 through 2.9.2, a lowest-role workspace member denied access to a tool by WorkspaceUserResourcePermission can still bind its identifier through tool_ids, skill_tool_ids, or mcp_tool_ids and execute it through the agent or workflow dispatch path. The dispatch path does not reapply the per-tool grant enforced by dedicated tool routes, and tool execution decrypts server-side init_params, allowing the caller to receive credentials carried by the denied tool. No fixed version is available as of this review.
CVE-2026-77518 is a missing authorization vulnerability in MaxKB, an open-source AI assistant for enterprise. In versions 2.10.2-lts and earlier, a normal workspace user can access hidden tools of other users within the same workspace by knowing the tool's MCP tool_id. This occurs because the tool-detail route lacks per-resource authorization checks, unlike the list route. The exposed tool details include Tool.code, which may contain sensitive MCP server configuration and headers. Additionally, an attacker who can create or edit workflows can misuse another user's MCP configuration by referencing their tool_id in a workflow node without permission verification. No fix is currently available for this issue.
CKAN MCP Server is a tool for querying CKAN open data portals. Prior to version 0.4.108, the SSRF guard `validateServerUrl` (added for CVE-2026-33060, extended for CVE-2026-53509) validates only the hostname string and never resolves DNS. Any caller-supplied `server_url` whose hostname *resolves* to an internal address passes the guard, so the server issues requests to loopback and cloud metadata (`169.254.169.254`). This is a third bypass of the same guard, and it reaches IMDS — strictly more than CVE-2026-53509, which only reached loopback. Version 0.4.108 contains an updated fix.
Fluent Bit is a fast and lightweight logs, metrics, and traces processor for Linux, BSD, macOS, and Windows. From 0.11.0 until 5.0.8, plugins/out_forward/forward.c secure_forward_pong copies the server-controlled PONG[2] reason into the 32-byte stack buffer msg with memcpy without checking its MessagePack type or length. An attacker who controls or can impersonate an out_forward Secure Forward destination configured with Shared_Key or Empty_Shared_Key can send an oversized reason during the first handshake and overwrite stack control data. Protected builds reliably terminate, while builds without a stack canary or with a disclosure can allow remote code execution as the Fluent Bit process user. When the opt-in --supervisor mode is used, fork-only respawns preserve the canary and address layout, allowing repeated crash-or-survive probes to support code execution on a hardened build; ordinary exec-based or service-manager restarts do not preserve that state. This issue is fixed in version 5.0.8.
CVE-2026-55071 is a high-severity code injection vulnerability in SepineTam's mcp-for-stata prior to version 1.19.0. The ado_package_install MCP tool concatenates user input directly into Stata commands without sanitization, allowing attackers to inject arbitrary Stata commands including shell escapes. This results in remote code execution at the OS level under the Stata-MCP server account. The vulnerability is present by default with no special configuration needed. It has been fixed in version 1.19.0.
AnyIO's TLSStream uses IDNA 2003 encoding for internationalized domain names, which can enable TLS certificate spoofing. An attacker who hijacks and redirects a TLS connection to a malicious server can obtain a legitimate certificate for the IDNA 2003 encoded domain and present it to the client, causing the client to accept the spoofed certificate. This vulnerability affects versions from 0.8.2 up to but not including 0.12.0_4. A fix is available in version 0.12.0_4. As a workaround, host names can be encoded using the 'idna' package before connecting.
CVE-2026-93965 is a command injection vulnerability in aiyiyi121 SxDevOps versions 1.0 and 1.1. The flaw exists in the subprocess.Popen function within the backend/aiops/services.py file of the MCP STDIO Server Management component. This vulnerability allows remote attackers to manipulate the argument endpoint_or_command, leading to command injection. The vendor has released a patch to fix this issue.
Obot versions up to 0.22.1 have a server-side request forgery (SSRF) vulnerability where privileged users can register remote MCP server URLs that Obot fetches without proper destination validation. This allows requests to internal network services and the cloud metadata endpoint (169.254.169.254), potentially exposing cloud IAM credentials. The vulnerability is fixed in version 0.23.0 by enforcing outbound egress restrictions on loopback, link-local, private IP ranges, and IPv6 ULA addresses.
Obot versions up to 0.22.1 with registry authentication enabled do not properly enforce authentication on the MCP Registry API endpoints under /v0.1/*. This allows unauthenticated users to read registry server metadata such as names, descriptions, repository URLs, icons, and connect URLs. No secret credentials or environment variables are exposed, and connecting to servers remains protected by authentication. The issue is fixed in version 0.23.0 and later.
Process Compose is a scheduler and orchestrator for non-containerized applications. Prior to 1.120.0, the MCP SSE listener in src/mcp/server.go accepts browser-origin requests to /sse and the returned message endpoint without validating the Host header, validating the Origin header, or authenticating the caller. When MCP SSE is enabled, a malicious website can use DNS rebinding to reach the loopback listener and issue MCP requests. If expose_control_tools is enabled, the attacker can enumerate process state, read or search logs, truncate logs, and start, stop, restart, or scale local processes; configured user-defined tools can expose additional commands and output. The Gin REST API token middleware does not protect this separately started MCP listener. This issue is fixed in version 1.120.0.
ToolHive is a utility designed to simplify the deployment and management of Model Context Protocol servers. Prior to ToolHive CLI 0.30.1 and ToolHive Studio 0.38.0, locally run MCP server containers use the default network permission profile without network isolation, permitting access to host.docker.internal while ToolHive API and MCP proxy endpoints are reachable without authentication. A malicious or compromised MCP server can use the Docker gateway to contact host-local services, other ToolHive-managed MCP proxies, or the ToolHive control plane without escaping the container. This access can expose data and logs, invoke sibling MCP tools, alter process or workload state, and disrupt services. ToolHive Studio additionally sends network_isolation as false and overrides the backend's secure isolation default. This issue is fixed in ToolHive CLI 0.30.1 and ToolHive Studio 0.38.0.
CVE-2026-50158 is a high-severity vulnerability in the eat-pray-ai yutu toolkit prior to version 0.10.9. It involves external control of a file name or path in the caption-download MCP tool, allowing an attacker with access to invoke caption-download to write files outside the intended confinement boundary. This can lead to overwriting critical files and potentially persistent code execution or denial of service. The vulnerability requires local HTTP client access when the MCP server runs without authentication and valid service credentials for live caption retrieval. The issue is fixed in version 0.10.9.
MKP is a Model Context Protocol server for Kubernetes. Prior to 0.4.1, cmd/server/main.go exposes the default HTTP endpoint and pkg/mcp/server.go registers the unauthenticated get_resource tool, which accepts attacker-controlled limitBytes and tailLines values for the pods logs subresource. buildPodLogOpts() in pkg/k8s/subresource.go parses those values as unbounded int64 parameters, and defaultGetPodLogs() copies the returned Kubernetes log stream through io.Copy into an in-memory bytes.Buffer without an application-side cap. A remote attacker who can reach the default port 8080 MCP endpoint and select a pod with sufficiently large accumulated logs can send one tools/call request that causes large allocations and additional response copies, while the request-frequency limiter does not constrain per-request volume. This can exhaust process memory, terminate the MKP server, and deny the MCP service; observed testing showed more than one GiB of RSS growth while handling a 128 MiB requested stream. This issue is fixed in version 0.4.1.
Obsidian Web MCP is a secure remote MCP server for Obsidian vaults. Prior to 0.2.0, /oauth/authorize issues an authorization code without a login, consent, or session check, and /oauth/token can exchange that code for the static VAULT_MCP_TOKEN without authenticating a client. An unauthenticated remote caller who can reach the intended tunnel deployment can therefore call /mcp and use vault_read, vault_write, vault_search, vault_list, vault_move, and vault_delete against the entire vault. Optional PKCE does not prevent an attacker-initiated flow, and unauthenticated /oauth/register also exposes a client_credentials path by returning the configured VAULT_OAUTH_CLIENT_SECRET. This issue is fixed in version 0.2.0.
CVE-2026-54504 is a vulnerability in andrea9293's mcp-documentation-server versions 1.13.0 up to but not including 1.13.1. The server's Web UI binds its document-management API to all network interfaces without authentication, allowing network-reachable attackers to access, modify, and delete documents without credentials. This issue is fixed in version 1.13.1.
NetLicensing MCP Server is a natural-language interface that enables agentic applications to manage the software-licensing lifecycle in Labs64 NetLicensing. Prior to 0.1.6, network-reachable HTTP transport requests to /mcp that omit x-netlicensing-api-key, Authorization: Bearer, and the apikey query parameter pass through ApiKeyMiddleware in src/netlicensing_mcp/server.py without authentication. The downstream api_key_ctx in src/netlicensing_mcp/client.py then falls back to the operator's NETLICENSING_API_KEY and authenticates upstream NetLicensing REST API calls under the operator account. An unauthenticated attacker can invoke MCP tools to enumerate products, licenses, licensees, and transactions, create or modify licensing objects, perform validations, and execute destructive delete operations. The issue affects HTTP deployments configured with a server-side key and does not require user interaction. This issue is fixed in version 0.1.6.
RMCP is an official Rust SDK for the Model Context Protocol. Prior to 2.0.0, the rmcp crate's OAuth implementation in crates/rmcp/src/transport/auth.rs omits the RFC 9728 resource field from ResourceServerMetadata and allows discover_oauth_server_via_resource_metadata to use protected-resource metadata without confirming that the returned resource identifier exactly matches the configured MCP server. A malicious MCP server can publish metadata for a different legitimate MCP resource and its authorization server, causing a victim who connects and completes the authorization flow to obtain a legitimate access token that the client subsequently sends to the malicious server. The attacker can capture the token and impersonate the victim against the legitimate MCP resource within the token's granted scopes. This issue is fixed in version 2.0.0.
RMCP is an official Rust SDK for the Model Context Protocol. Prior to 2.0.0, the rmcp crate's stateful Streamable HTTP server in crates/rmcp/src/transport/streamable_http_server/tower.rs allows an unauthenticated client to send a well-formed JSON-RPC POST that is not an initialization request, or an initialization request with a mismatched protocol header, causing StreamableHttpService::handle_post to call LocalSessionManager.create_session before validating the message. An early validation failure returns without removing the inserted LocalSessionHandle from LocalSessionManager.sessions, permanently retaining session and channel state for the server process lifetime. Repeated requests can grow the shared session table without bound, degrade legitimate-client latency through lock contention, exhaust memory, and terminate the server. This issue is fixed in version 2.0.0.
@zereight/mcp-gitlab version prior to 2.1.30 contains multiple security flaws that bypass key safety controls such as read-only mode, project allow-list, and transport authentication. These include a GraphQL mutation bypass that allows unauthorized writes, unauthenticated access to HTTP and SSE transports, a session exhaustion denial-of-service, and exposure of CI job traces that can be manipulated by attackers. These vulnerabilities enable attackers with access to the MCP server port or a semi-trusted client to perform unauthorized GitLab writes, access GitLab credentials without authentication, and cause denial of service.