AWS repairs an MCP server bug that could leak a database token
Version 0.0.19 repairs the bug, and users that had the bug must also replace their tokens.
On 16 November 2026, AWS wrote a bulletin on CVE-2026-126681. This is a bug in its open source Timestream for InfluxDB MCP server. MCP (Model Context Protocol) is a standard that lets AI assistants use external tools. The server lets an assistant control InfluxDB and read its data. The server did not limit the endpoints that it connects to.
How it works
An attacker with no login can use prompt injection. This is text that gives an AI an incorrect task. The attacker puts a made InfluxDB URL in the context of the AI client. The server then can send its API token to an endpoint that the attacker controls.
The repair
AWS tells users to install version 0.0.19. It also tells them to patch forked code and to replace each API token that the server used. AWS gave no test results, no count of users and no sign that an attacker used the bug.
What is not repaired
No workaround repairs the bug fully. Until you install the new version, do these steps:
- Use a token with the minimum permissions that the work must have.
- Do not use the server with clients that read untrusted content.
Insight: AWS MCP servers keep failing at the same point
Since September 2026, AWS wrote bulletins on bugs in its Postgres, API and gateway MCP servers. This bulletin is the fourth one. Each bug let something that a user or an attacker sent make a server do more than it must. This reads as one weak point: the MCP server that holds the credentials, not the AI model. Leaders must examine MCP servers as they examine a system with credentials.
Sources
- CVE-2026-126681Improper restriction of intended endpoints in AWS Labs Timestream for InfluxDB MCP Server
Posted