AAK-MCP-CONTEXTFORGE-CVE-2026-110279-001: IBM ContextForge MCP Gateway 0.8.0 to 1.1.0: X-Upstream-Authorization replaces the gateway's upstream credential, identity headers skip the per-gateway setting, jq guard DoS
- Severity
- HIGH
- Category
- Supply Chain (
supply-chain) - Scanner
agent_audit_kit/scanners/container_image_pins.py,agent_audit_kit/scanners/mcp_cve_pins_2026_07.py- Auto-fixable
- No.
- SARIF name
ContextForgeUpstreamCredentialOverride
What it detects
IBM ContextForge MCP Gateway (mcp-contextforge-gateway on PyPI, and the ghcr.io/ibm/mcp-context-forge image) up to 1.1.0 carries three flaws that 1.1.1 fixes. A client's X-Upstream-Authorization header is always renamed to Authorization on the request the gateway sends upstream, after the gateway has set the upstream credential it manages (a stored per-user OAuth token, a client-credentials token, or the gateway's bearer, basic or custom auth headers), so an authenticated user replaces that credential with another one and the upstream server acts under it while the gateway's audit trail names only the user (CVE-2026-110279, CVSS 3.1 8.1 from IBM, CWE-287). The rename arrives in 0.8.0, always on and ahead of the ENABLE_HEADER_PASSTHROUGH check, and 0.7.0 has no such header; 1.1.1 (#7341) drops it for a gateway whose auth type is oauth, bearer, basic or authheaders. With identity propagation on, two tool invocation paths and a resource read path build the identity headers without the gateway, so a server whose per-gateway setting turns propagation off still receives the user's id, email, groups and teams, and, where claims are signed, a signature over the user id and email alone, the same on every request (CVE-2026-110278, CVSS 3.1 7.4, CWE-294). Identity propagation arrives in 1.0.0 (0.9.0 has none), and 1.1.1 (#7329) passes the gateway's setting on those paths. The jq filter guard that 1.0.8 added omits halt, halt_error, builtins and the unbounded-iteration built-ins (while, until, recurse, recurse_down, foreach, repeat), so an authenticated user's jsonpath_filter ends a gateway worker or keeps it busy (CVE-2026-110189, CVSS 3.1 6.5, CWE-184); 1.0.8 to 1.1.0 carry the same guard, and 1.1.1 adds them. Checked in the PyPI wheels of 0.7.0, 0.8.0, 0.9.0, 1.0.0, 1.0.7.post20260921, 1.0.8, 1.0.11, 1.1.0 and 1.1.1. IBM's bulletins name 1.0.0 through 1.1.0 for the first two and 1.0.8 through 1.1.0 for the third, and 1.1.1 as the fix for all three.
How to fix
Upgrade mcp-contextforge-gateway to 1.1.1 or later and pin it, or run the ghcr.io/ibm/mcp-context-forge image at v1.1.1 or later. Until you can, remove the X-Upstream-Authorization header from client requests at the reverse proxy in front of the gateway's MCP endpoints, set IDENTITY_PROPAGATION_ENABLED=false where any upstream server must not receive user identity, and keep the gateway off networks you do not trust. 1.1.1 still signs identity claims over the user id and email alone, so an upstream server should not treat the signature as proof that a request is fresh.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-MCP-CONTEXTFORGE-CVE-2026-110279-001
Limitations
Reads mcp-contextforge-gateway from 0.8.0 below 1.1.1 where a requirements*.txt entry at any depth, or at the project root a pyproject.toml, Pipfile, lockfile or MCP config entry, states the version, judging a range by the versions it admits. latest is 1.1.1, fixed, so a reference that states no version is not reported. The 1.0.0 betas and release candidates read as 1.0.0 and 1.0.7.post20260921 as 1.0.7, all inside the range. IBM's bulletins start at 1.0.0, its supported line; the range starts at 0.8.0, where the wheels show the header rename, so 0.8.0 and 0.9.0 are reported under this rule's identity and jq records too, and 1.0.0 to 1.0.7 under its jq record, before those features arrived. It reads the version, not the deployment's settings (which auth type a gateway uses, whether identity propagation or jq filters are in use, what a proxy in front strips), so a deployment that uses none of the affected features is reported too. It also reads ghcr.io/ibm/mcp-context-forge tags from 0.8.0 below 1.1.1, with or without the v, in compose files, Kubernetes manifests, Helm values, Dockerfiles and MCP configs; latest was the v1.1.1 digest on 2026-11-14, fixed, so an untagged or latest reference is not reported. 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-2026-110279, CVE-2026-110278, CVE-2026-110189
- OWASP MCP Top 10: MCP07:2025 (Insufficient Authentication & Authorization), MCP01:2025 (Token Mismanagement & Secret Exposure), MCP10:2025 (Context Injection & Over-Sharing), MCP04:2025 (Software Supply Chain Attacks & Dependency Tampering)
- OWASP Agentic Top 10: ASI03 (Identity and Privilege Abuse), ASI04 (Agentic Supply Chain Vulnerabilities), ASI08 (Cascading Failures)
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-SUPPLY-08 (MCP-Specific Vuln)
- CSA AI Controls Matrix (AICM): IAM-01, 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-CONTEXTFORGE-CVE-2026-110279-001.md: the page then shows that text, with the registry reference appended under it.