Skip to content
awslabs/mcp·Indexed by this site · Originally publishedDate not given· Event →· Its place in the scan record →

Prompt injection can make the InfluxDB MCP server send tokens to a malicious URL

Improper restriction of intended endpoints in AWS Labs Timestream for InfluxDB MCP Server

AI summary

AWS disclosed an improper endpoint restriction vulnerability in its open-source Timestream for InfluxDB MCP Server, where prompt injection can lead to leakage of InfluxDB API tokens (CVE-2026-126681). The server lets MCP clients such as AI assistants manage Timestream for InfluxDB resources; a remote unauthenticated attacker can, through prompt injection, implant a crafted InfluxDB URL in the MCP client context, causing the server to connect to that URL and send the configured token to an attacker-controlled endpoint, thereby accessing customer InfluxDB data with that token's permissions. Versions 0.0.12 and above and before 0.0.19 are affected, and it has been fixed in 0.0.19. AWS recommends affected users upgrade and rotate InfluxDB API tokens; there is no workaround that fully resolves the issue, and before upgrading, least-privilege tokens should be used and pairing with MCP clients that process untrusted content should be avoided.

Read originalgithub.com