{"document":{"category":"csaf_vex","csaf_version":"2.0","title":"CVE-2026-48774: ProxySQL MCP run_sql_readonly executes side-effecting MySQL multi-statements despite read-only contract","publisher":{"category":"vendor","name":"HarborGuard Database","namespace":"https://database.harborguard.co"},"tracking":{"id":"CVE-2026-48774","status":"final","version":"1","initial_release_date":"2026-06-19T19:34:39.971Z","current_release_date":"2026-06-19T19:34:39.971Z","revision_history":[{"date":"2026-06-19T19:34:39.971Z","number":"1","summary":"Initial machine-readable export from HarborGuard."}]},"distribution":{"tlp":{"label":"WHITE"},"text":"Public CVE data; freely redistributable."},"notes":[{"category":"description","text":"ProxySQL is a proxy for MySQL and its forks, as well as PostgreSQL. In versions 3.0.0 through 3.0.8, ProxySQL's GenAI/MCP `run_sql_readonly` tool violates its documented read-only contract for MySQL targets. The tool validates only the full input string with a substring blacklist and first-keyword allowlist, but then executes the entire SQL string on a backend connection created with `CLIENT_MULTI_STATEMENTS`. As a result, a caller can submit a read-only first statement followed by a side-effecting second statement, such as `SELECT 1; RENAME TABLE ...`. The validator accepts the payload because it starts with `SELECT` and because side-effecting MySQL statements such as `RENAME TABLE`, `SET`, `RESET`, `LOCK TABLES`, and `KILL` are not rejected by the blacklist. In a live MCP runtime test, the `/mcp/query` endpoint accepted a `run_sql_readonly` request. The MCP response reported success for the first `SELECT`, and direct backend verification showed that the table had actually been renamed. This violates the endpoint's read-only security contract and lets an MCP caller perform backend writes or administrative SQL, limited by the configured MCP target account's database privileges. Version 3.0.9 contains a fix. Other operator mitigations include: keeping MCP disabled unless required; setting a non-empty `mcp-query_endpoint_auth` token before exposing `/mcp/query`; restricting MCP listener network exposure; configuring MCP backend target credentials as database-level read-only users; and adding temporary MCP query rules to block obvious multi-statement patterns.","title":"CVE description"}],"references":[{"category":"self","summary":"CVE-2026-48774 on HarborGuard Database","url":"https://database.harborguard.co/cve/CVE-2026-48774"},{"category":"external","summary":"CVE Record","url":"https://www.cve.org/CVERecord?id=CVE-2026-48774"},{"category":"external","summary":"https://github.com/sysown/proxysql/security/advisories/GHSA-7wh6-2vcc-gcm4","url":"https://github.com/sysown/proxysql/security/advisories/GHSA-7wh6-2vcc-gcm4"},{"category":"external","summary":"https://github.com/sysown/proxysql/commit/e32b7fd50c7c234ea628e392e621e09a2a919e08","url":"https://github.com/sysown/proxysql/commit/e32b7fd50c7c234ea628e392e621e09a2a919e08"}]},"product_tree":{"branches":[{"category":"vendor","name":"sysown","branches":[{"category":"product_name","name":"proxysql","branches":[{"category":"product_version","name":">= 3.0.6, < 3.0.9","product":{"name":"sysown proxysql >= 3.0.6, < 3.0.9","product_id":"CSAFPID-1","product_identification_helper":{"cpe":"cpe:2.3:a:sysown:proxysql:\\>\\=_3.0.6\\,_\\<_3.0.9:*:*:*:*:*:*:*"}}}]}]}]},"vulnerabilities":[{"cve":"CVE-2026-48774","title":"ProxySQL MCP run_sql_readonly executes side-effecting MySQL multi-statements despite read-only contract","notes":[{"category":"description","text":"ProxySQL is a proxy for MySQL and its forks, as well as PostgreSQL. In versions 3.0.0 through 3.0.8, ProxySQL's GenAI/MCP `run_sql_readonly` tool violates its documented read-only contract for MySQL targets. The tool validates only the full input string with a substring blacklist and first-keyword allowlist, but then executes the entire SQL string on a backend connection created with `CLIENT_MULTI_STATEMENTS`. As a result, a caller can submit a read-only first statement followed by a side-effecting second statement, such as `SELECT 1; RENAME TABLE ...`. The validator accepts the payload because it starts with `SELECT` and because side-effecting MySQL statements such as `RENAME TABLE`, `SET`, `RESET`, `LOCK TABLES`, and `KILL` are not rejected by the blacklist. In a live MCP runtime test, the `/mcp/query` endpoint accepted a `run_sql_readonly` request. The MCP response reported success for the first `SELECT`, and direct backend verification showed that the table had actually been renamed. This violates the endpoint's read-only security contract and lets an MCP caller perform backend writes or administrative SQL, limited by the configured MCP target account's database privileges. Version 3.0.9 contains a fix. Other operator mitigations include: keeping MCP disabled unless required; setting a non-empty `mcp-query_endpoint_auth` token before exposing `/mcp/query`; restricting MCP listener network exposure; configuring MCP backend target credentials as database-level read-only users; and adding temporary MCP query rules to block obvious multi-statement patterns.","title":"CVE description"}],"product_status":{"known_affected":["CSAFPID-1"]},"scores":[{"cvss_v3":{"version":"3.1","vectorString":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:H/A:N","baseScore":7.5,"baseSeverity":"HIGH"},"products":["CSAFPID-1"]}],"remediations":[{"category":"none_available","details":"No fixed version is published yet. Monitor the upstream advisory.","product_ids":["CSAFPID-1"]}]}]}