AI Agent News/stories/pydantic-ai-patches-medium-severity-mcp-telemetry-credential-leak-in-v1-107-8.md

Frameworks

Pydantic AI patches medium-severity MCP telemetry credential leak in v1.107.8

Pydantic AI v1.107.8 fixes a medium-severity issue in which MCP server URLs, authorization tokens and header values could appear in OpenTelemetry traces. The same fix ships in v2.58.0, and the maintainers advise restricting access to historical traces and rotating any exposed credentials.

Source: github.com/pydantic/pydantic-ai/releases/tag/v1.107.8

The fix covers provider-native MCP configuration, which could expose server URLs, authorization tokens and custom header values in OpenTelemetry request traces even when include_content is set to False. URL-based MCP capabilities could also leak a credential embedded in the URL-derived identifier. The advisory is tracked as GHSA-xv4x-g8mx-ch56 and was released on 19 October 2026.

The maintainers recommend restricting access to historical traces and rotating any credentials that may already have been exposed. Users of Temporal tracing have a caveat: URL-based MCP activity names were left unchanged to keep in-flight workflows running. After those workflows finish, the project suggests passing an explicit id= to MCP, though changing it renames the activities.

Why it matters if you build agents

If you run MCP servers through Pydantic AI with OpenTelemetry tracing, upgrade to v1.107.8 or v2.58.0 and review who can read existing traces. Treat any URL-embedded or header credentials that may have appeared in past traces as exposed and rotate them. Under Temporal tracing, expect activity renames if you change an MCP id.

Get stories like this every weekday

One short email with the day's agent tooling news.

Subscribe