AAK-MCP-REMOTE-CVE-2026-108716-001: mcp-remote from 0.8.0 launched with --device-code: client secret and tokens over plain HTTP (no fixed release)
- Severity
- MEDIUM
- Category
- Supply Chain (
supply-chain) - Scanner
agent_audit_kit/scanners/mcp_cve_pins_2026_07.py- Auto-fixable
- No.
- SARIF name
McpRemoteDeviceCodeCleartext
What it detects
mcp-remote's device-code sign-in (--device-code, from 0.8.0) posts the OAuth client secret to the device-authorization and token endpoints the server's metadata names, and takes the access and refresh tokens back, without requiring https. When a discovered endpoint is a non-loopback http:// URL, an on-path attacker captures the secret and the tokens (CVE-2026-108716, CVSS 3.1 5.3, CWE-319). Checked in the source: authorizeWithDeviceCode arrives in 0.8.0, runs only when the launch passes --device-code, and calls fetch on both endpoints with no scheme check; src/lib/device-authorization.ts is unchanged from 0.14.3 to 0.15.1, the newest release, so there is no fixed release.
How to fix
There is no fixed release. Sign in without --device-code, or check that the server's OAuth metadata names https device-authorization and token endpoints before you use it, and rotate a client secret that may have crossed plain HTTP.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-MCP-REMOTE-CVE-2026-108716-001
Limitations
Reports an MCP config or package.json script that launches mcp-remote with --device-code, at a stated version from 0.8.0 or with none, since the newest release is affected. A launch without the flag is not reported, and neither is a dependency entry or a lockfile, which record no arguments. A launch in .cursor/mcp.json or .vscode/mcp.json is not read. It cannot see the endpoints the server will advertise, so a server whose endpoints are https is reported too. A file over 2 MB, or a lockfile over 32 MiB, is not read. A declaration nested more than 64 levels deep in package.json, a TOML manifest or an MCP config (an MCP server more than 62) is still judged, and reported without a line; a lockfile entry keeps its line at any depth.
References
- CVE: CVE-2026-108716
- OWASP MCP Top 10: MCP01:2025 (Token Mismanagement & Secret Exposure), MCP04:2025 (Software Supply Chain Attacks & Dependency Tampering)
- OWASP Agentic Top 10: ASI03 (Identity and Privilege Abuse), ASI04 (Agentic Supply Chain Vulnerabilities)
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-TOKEN-06 (Generic Secret Exposure)
- CSA AI Controls Matrix (AICM): CEK-08, IVS-04, STA-08
All rules. This page is generated from the rule registry when the documentation is built; to change it, change the rule in agent_audit_kit/rules/builtin.py. To give the rule a hand-written write-up, add docs/rules/AAK-MCP-REMOTE-CVE-2026-108716-001.md: the page then shows that text, with the registry reference appended under it.