GHSA-MGMV-R532-PG36
A credential in an MCP server URL path could reach chat frontends through MCP Apps events
Texto original en inglés · Ver en español
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N
Fuente: GitHub
Ciclo de vida de la vulnerabilidad
| Fecha | Fase | Evento | Detalle |
|---|---|---|---|
| Remediación | Parche 2.60.0 | pydantic-ai · -12 d desde divulgación | |
| Divulgación | Publicada | GHSA-MGMV-R532-PG36 divulgada · CVSS 6.5 (GitHub) | |
| Monitoreo | Detectada | Ingresó al índice | |
| Monitoreo | Verificada | Última verificación contra las fuentes |
Summary
An `MCP` capability configured with a server URL and no explicit `id` derived its identifier from the URL's hostname and last path segment. The AG-UI and Vercel AI adapters send that identifier to the frontend with MCP Apps tool results, so a credential in the URL's last path segment could reach users of the application's chat frontend.
Details
You are affected if all of the following hold:
- You use the `MCP` capability with a server `url=`, run the server locally (the default on v2), and pass no `id=`.
- The server URL carries a credential in its last path segment.
- The server exposes MCP Apps tools, and you stream runs to a frontend with `AGUIAdapter` or `VercelAIAdapter`.
When the model calls an MCP Apps tool, the adapter's MCP Apps event carries the derived identifier as its server identifier. The same value is kept in message history dumped by those adapters. The assumed attacker is an authenticated user of the application's chat frontend, who can read their own event stream and can ask the agent to use the server's tools. That user obtains the application's credential for the MCP server.
An `MCPToolset` passed directly, and an `MCP` capability with an explicit `id=`, are not affected. GHSA-xv4x-g8mx-ch56 covered the same identifier in traces; the frontend channel remained exposed until 2.60.0, so upgrading to 2.58.0 or 2.59.0 did not address it. The v1 line does not forward MCP Apps results and is not affected.
Mitigation
Upgrade `pydantic-ai` and `pydantic-ai-slim` to 2.60.0 or later, where the derived identifier is the hostname only. Rotate MCP server credentials that were carried in the URL path, since they may have been streamed to frontend users or stored in dumped message history.
Upgrading changes the derived `id` of such a capability, which also names its durable operations (Temporal activities, DBOS steps, Prefect tasks). If you run one under durable execution, let in-flight workflows complete before upgrading.
Desglose de métricas
Rangos afectados
Estado del parche por paquete
| Paquete / producto | Estado | Corregida en | Desde divulgación | Última versión |
|---|---|---|---|---|
pypipydantic-ai |
Corregida | 2.60.0 | -12 d | 2.68.0 no afectada |
pypipydantic-ai-slim |
Corregida | 2.60.0 | -12 d | 2.71.0 no afectada |
2.60.0Arsenal público
Sin exploits, módulos ni PoC públicos conocidos a la fecha de verificación.
https://github.com/pydantic/pydantic-ai/security/advisories/GHSA-mgmv-r532-pg36Las referencias externas se muestran para copiar: cve.llama.ec no lo redirige a otros sitios.
