AWS Labs MCP servers: prompt injection can make aws-pricing-mcp-server write files anywhere and timestream-for-influxdb-mcp-server send its InfluxDB token to an attacker (CVE-2026-126816, CVE-2026-126681)
AWS published two more advisories for its open-source awslabs MCP servers on 16 November 2026. In awslabs.aws-pricing-mcp-server before 1.0.31 the generate_cost_report tool, wrongly annotated read-only, wrote reports to any path the agent supplied. In awslabs.timestream-for-influxdb-mcp-server 0.0.12 up to 0.0.19 the server connected to whatever InfluxDB URL appeared in the conversation and sent its configured API token there. Both are reachable through content the agent reads.
What it is
CVE-2026-126816 / GHSA-mqm3-w68j-75mw (AWS bulletin 2026-184): absolute paths or ../ in the cost report's output_file let attacker-influenced content create or overwrite files outside the output directory, with attacker-supplied field values in the report. Because the tool was marked readOnlyHint, clients that auto-approve read-only tools ran it without asking. Affected: aws-pricing-mcp-server < 1.0.31 and the deprecated awslabs.cost-analysis-mcp-server <= 1.0.5 (no fix coming for that one).
1.0.31 confines output_file to a base directory (AWS_PRICING_MCP_OUTPUT_DIR, default the working directory), rejects traversal and escaping symlinks, and corrects readOnlyHint to false.
CVE-2026-126681 / GHSA-2f9r-g4f3-wc2j (AWS bulletin 2026-185): the connection configuration did not restrict endpoints, so a remote unauthenticated actor could use prompt injection to introduce a crafted InfluxDB URL and receive the server's InfluxDB API token, then use it with that token's permissions. Fixed in 0.0.19.
AWS says no managed AWS service is affected; impact is limited to the machine running the MCP server and the configured InfluxDB token.
How to check
- Run `pip show awslabs.aws-pricing-mcp-server awslabs.cost-analysis-mcp-server awslabs.timestream-for-influxdb-mcp-server` (or check uvx launch lines in your MCP client config).
- Check whether generate_cost_report is in any auto-approve list.
How to fix
- Upgrade to awslabs.aws-pricing-mcp-server 1.0.31+ (migrate off awslabs.cost-analysis-mcp-server) and awslabs.timestream-for-influxdb-mcp-server 0.0.19+.
- Rotate InfluxDB API tokens that were configured for the Timestream server, and use least-privilege tokens.
- Until upgraded, remove generate_cost_report from auto-approve and avoid using these servers in sessions that read untrusted content.
pip show awslabs.aws-pricing-mcp-server awslabs.cost-analysis-mcp-server awslabs.timestream-for-influxdb-mcp-server 2>/dev/null | grep -E '^(Name|Version)' Sources
- criticalMalicious npm package @guicai777/mcp-doctor posed as an MCP diagnostic tool and dropped a hidden Windows executable, with a README telling users to run mshta (MAL-2026-20295)
- criticalCVE-2026-77359: Jupyter MCP Server's /api/connect and /api/stop routes skipped the token and Origin checks, so a website or network caller could repoint it to an attacker's Jupyter backend (fixed 1.0.3)
- criticalMalicious npm packages posing as AI SDKs and MCP servers: character-ai-sdk, char-ai-sdk, @mibbp/agent-observer, mrnima-openai-chatgpt-bot and @fdcasj_dfew/md-doc-mcp
- criticalCVE-2026-76807: Doist's official Todoist MCP HTTP server let any website use DNS rebinding to read, create and delete your Todoist data (fixed 10.3.3)