DBHub vulnerabilities

SQL injection vulnerability in dbhub v1.2.0 in the /utils/sql-parser.ts component. Attackers can access sensitive database information via crafted SQL statements.

DBHub, a database MCP server, in versions prior to 0.22.5 exposes an unauthenticated HTTP MCP endpoint when started in HTTP transport mode. Its protection based on comparing Origin and Host headers does not prevent DNS rebinding, allowing a malicious website to invoke MCP tools from the victim's browser. Depending on configuration, this can lead to reading, enumerating, and potentially writing database contents.

In DBHub prior to version 0.22.6, read-only SQL validation can be bypassed on the MySQL and MariaDB connectors. A double dash followed by a non-whitespace character can hide a subsequent write or destructive DDL statement from the validator while the database interprets the dashes as minus operators and executes both statements. Anyone able to influence the assistant's SQL, including through prompt injection, can bypass read-only mode to modify or drop database data.

DBHub prior to version 0.22.6 does not enforce read-only mode for the execute_sql tool even when readonly = true is set. The code that would enforce read-only is never executed, leaving only a simple classifier that checks the first keyword of each statement. This allows sequence tampering, arbitrary file writes, host file reads, and remote code execution. The HTTP transport is unauthenticated and binds to 0.0.0.0 by default.