dbt-mcp is a Model Context Protocol server for interacting with dbt. Prior to 1.20.0, the local OAuth helper in src/dbt_mcp/oauth/fastapi_app.py exposes GET /dbt_platform_context without authentication or Host validation after a user completes the dbt Platform OAuth flow. The endpoint returns the full DbtPlatformContext, including access_token and refresh_token values persisted by the context manager, to any process that can reach 127.0.0.1:6785. The absence of TrustedHostMiddleware allows a remote attacker to use DNS rebinding against a victim's browser because the helper accepts arbitrary Host headers, while a co-located process can request the endpoint directly. The stolen tokens provide immediate dbt Platform API access as the victim and persistent access through the refresh token, allowing access to or modification of projects, jobs, environment secrets, and related account data. This issue is fixed in version 1.20.0.
A vulnerability was detected in GH05TCREW PentestAgent up to cf882dabea3ed91cef016cdd115e5426315665a2. This vulnerability affects the function run_task of the file interface/main.py of the component MCP HTTP Server. Performing a manipulation results in os command injection. The attack is possible to be carried out remotely. The exploit is now public and may be used. This product adopts a rolling release strategy to maintain continuous delivery. Therefore, version details for affected or updated releases cannot be specified. The pull request to fix this issue awaits acceptance.
A campaign named Miasma exploited a vulnerability in the Model Context Protocol (MCP) tool configuration process to gain OS-level access to 73 GitHub repositories, including one owned by Microsoft Azure. The attack leverages the fact that MCP clients trust tool descriptions as executable instructions without re-prompting or sandboxing, allowing malicious edits to MCP config files to run with developer privileges. This attack class, known as MCP tool poisoning, was first documented in 2025 and escalated to real-world exploitation by mid-2026. The campaign abused the approval process tied to server names rather than content, enabling malicious commands to execute automatically after initial approval. OWASP ranks this attack third in its MCP Top 10 threats. Mitigations include rigorous code review of MCP config changes, pinning tool definitions, scoping privileges per tool, maintaining allowlists of MCP servers, and sandboxing agent execution.
MCPHub before 1.0.32 contains an authentication bypass vulnerability in its embedded OAuth 2.0 authorization server where client authentication is disabled by default and PKCE enforcement is optional. Attackers who obtain an authorization code through interception can redeem it for access tokens without providing a client secret or PKCE verifier, gaining access to victim accounts and their privileges.
MySQL MCP Server is a Model Context Protocol server that enables secure interaction with MySQL databases. Prior to 0.4.2, setting MCP_TRANSPORT=sse causes src/mysql_mcp_server/server.py to construct SseServerTransport without security_settings or enable_dns_rebinding_protection, while the Starlette routes /, /sse, and /messages/ have no authentication and the service binds to 0.0.0.0 by default. A network attacker can directly reach execute_sql, or can use DNS rebinding to make a victim's browser relay same-origin requests to a locally bound service, and supply a query that reaches cursor.execute(query). This allows unauthenticated disclosure and modification of the configured database; when the MySQL account has FILE privileges, the same access can read or write server files and may enable code execution. The default stdio transport is not affected. This issue is fixed in 0.4.2.
IBM Langflow OSS 1.0.0 through 1.11.5 could allow a remote authenticated attacker to execute arbitrary commands due to improper validation of command-line arguments in the MCP stdio server configuration.
IBM Langflow OSS 1.0.0 through 1.11.5 allows an authenticated non-administrative user could execute arbitrary operating system commands on the server at the privilege level of the application process by constructing a flow with an MCP Tools component configured to use a local stdio subprocess transport. This bypasses both the LANGFLOW_CUSTOM_COMPONENT_ADMIN_ONLY and LANGFLOW_BLOCK_CODE_INTERPRETER_COMPONENTS server-side controls intended to prevent exactly this class of access. Successful exploitation could lead to arbitrary command execution, sensitive data exposure (including credentials from the process environment), file system modification, and lateral movement to services reachable from the server.
The functype-mcp-server contains a high-severity remote code execution vulnerability (CVE-2026-59176) in its set_functype_version tool. This tool accepts an unsanitized version string that is directly interpolated into an npm package specifier and installed via pnpm without validation. An attacker can exploit this by supplying a malicious package alias, causing the server to install and dynamically import attacker-controlled code, resulting in full RCE with server process privileges. The vulnerability affects versions prior to 1.4.4 and is patched in version 1.4.4.
An OS command injection weakness in the read-only enforcement of the SQL validation component in Amazon awslabs postgres-mcp-server before 1.1.7 might allow an unauthenticated actor to execute operating system commands on the host of a self-managed PostgreSQL server by placing a crafted COPY ... TO PROGRAM statement into content that is processed when an authenticated user interacts with the MCP server in its default read-only mode.
To remediate this issue, users should upgrade to version 1.1.7 or later.
CVE-2026-85788 is a medium severity vulnerability in AWS Labs MySQL MCP Server where an incomplete list of disallowed inputs in the mutable SQL detector allows certain SQL inline comments to bypass read-only enforcement. This can enable context-dependent actors to reach file-read and file-write SQL sinks. The issue is fixed in version 1.0.23.
eprosima Fast DDS is a C++ implementation of the DDS (Data Distribution Service) standard of the OMG (Object Management Group). Versions prior to 2.6.12, 2.14.6, 3.2.4, 3.3.1, and 3.4.2 have a remotely triggerable Out-of-Bounds Read while processing RTPS `DATA_FRAG` submessages. An attacker can craft a `DATA_FRAG` with a large `sampleSize` but a small actual payload, and set `fragmentsInSubmessage` such that the receiver treats the packet as the LAST fragment**. In this LAST-fragment path, Fast-DDS computes `incoming_length` based on `sampleSize` and calls `memcpy()` without validating `incoming_data.length >= incoming_length`. As a result, `CacheChange_t::add_fragments()` reads past the received UDP datagram buffer and into adjacent heap memory, copying those bytes into the reassembly buffer. In a Discovery Server deployment, the resulting `CacheChange_t` can be relayed to other participants, meaning that a newly joining participant may receive leaked heap memory (e.g., pointer values that could aid ASLR bypass). Versions 2.6.12, 2.14.6, 3.2.4, 3.3.1, and 3.4.2 fix the issue.
knowns versions before 0.30.0 fail to validate filesystem paths in MCP tool arguments, allowing attackers to read, create, overwrite and delete files outside the project directory. Attackers can supply path arguments containing directory traversal sequences to access arbitrary Markdown files accessible to the server process.
This security update for mcphost addresses two vulnerabilities: CVE-2026-41178, which involves denial of service via oversized inputs in baggage parsing, and CVE-2026-81092, which allows resource exposure due to lack of Host header checks in HTTP transports. The update includes upgrades to dependencies github.com/mark3labs/mcp-go/server (v0.56.0) and go.opentelemetry.io/otel (1.44.0).
Rowboat through 0.9.1 fails to validate custom MCP server and webhook URLs, allowing authenticated users to configure arbitrary destinations. Attackers can point these URLs at internal services and cloud metadata endpoints to perform server-side request forgery and enumerate internal network topology.
CVE-2026-85787 is a high-severity vulnerability in Amazon's awslabs postgres-mcp-server before version 1.1.7. It involves an incomplete list of disallowed inputs in the SQL validation component, which may allow an unauthenticated attacker to modify data beyond intended read-only permissions by injecting crafted SQL when an authenticated user interacts with the server. The issue is fixed in version 1.1.7.
Improper neutralization of special elements used in a template engine in the CDK generator in Amazon awslabs.dynamodb-mcp-server before 2.1.6 might allow a context-dependent actor to execute arbitrary code on the host that deploys the generated application via crafted table, index, or attribute names in a data model file.
OGX (formerly Llama Stack, affected at commit fbe8e0f) contains an unauthenticated server-side request forgery vulnerability in the OpenAI-compatible POST /v1/responses endpoint. MCP tool definitions accept a server_url parameter (along with headers and authorization values) that is fetched server-side without destination validation; the existing validate_url_not_private() guard used for other URL inputs is not applied to server_url. On the default starter configuration, which runs without authentication, a remote unauthenticated attacker can cause the server to open connections to arbitrary internal addresses (including cloud metadata endpoints such as http://169.254.169.254/) and forward attacker-supplied headers and bearer tokens to those destinations.
cli-mcp-server 0.2.5 contains a command allowlist bypass vulnerability in the _validate_command_with_operators function when ALLOW_SHELL_OPERATORS is enabled. Attackers can use shell command substitution syntax like $(...) or backticks to execute non-allowlisted commands that bypass the ALLOWED_COMMANDS validation check.
git-mcp-server 2.15.1 contains an argument injection vulnerability in the ref and object parameters of git_log, git_diff, and git_show tools that lack leading-dash validation. Attackers can inject git command-line options like --output= to write files outside the repository to arbitrary paths accessible by the process.