AI Stack Alert High CVE-2026-102610 For AI Eng, AppSec, SOC / IR, CISO AI-assisted
fast-agent serve (before 0.7.21): unauthenticated /mcp endpoint, optional remote shell (CVE-2026-102610)
Prior to version 0.7.21, fast-agent serve binds its HTTP transport to 0.0.0.0 on port 8000 with no authentication. When started with --shell, an unauthenticated remote client can cause operating-system commands to run as the server user. The NVD entry lists a CVSS base score of 6.4.
Why it matters
Any network client that can reach the server's port 8000 can use the exposed agent tool. With --shell enabled, this gives remote command execution as the server user. Without --shell, an attacker can still use the operator's model and configured MCP tools at the operator's cost and access.
Do this
- Upgrade fast-agent to version 0.7.21 or later.
- Until patched, do not run fast-agent serve with --shell.
- Restrict access to port 8000 with a host firewall or network ACL, or bind the server to 127.0.0.1 instead of 0.0.0.0.
- Check for listening sockets on 0.0.0.0:8000 on hosts running fast-agent, and review any servers found exposed beyond trusted networks.
- On hosts where --shell was enabled, review process and shell logs for unexpected child processes spawned by the fast-agent server user.
Exploit Clock
Exploitation unlikely (for now)
- EPSS 0.29%
Signals
- NoCISA KEV
- 6.4CVSS
- 0.29%EPSS
- NoPublic PoC
- 36/100Priority
- 80%Confidence
Evidence
“Prior to 0.7.21, fast-agent serve uses HTTP transport bound to 0.0.0.0 on port 8000 without authentication, allowing any network client that can reach the /mcp endpoint to invoke the exposed agent tool.”
“an unauthenticated remote client can cause operating-system commands to run as the server user”
“without --shell, the client can still consume the operator's model and configured MCP tools”
Sources
Affected / related
fast-agent, fast-agent serve, Model Context Protocol (MCP) HTTP transport
Published 2026-11-06 21:17 UTC · updated 2026-11-08 02:26 UTC · AI-assisted summary, checked against quoted sources · How we score