AAK-MCP-ATLASSIAN-CVE-2026-73498-001: mcp-atlassian below 0.22.0 (confluence_upload_attachment arbitrary file read)
- Severity
- HIGH
- Category
- Supply Chain (
supply-chain) - Scanner
agent_audit_kit/scanners/container_image_pins.py,agent_audit_kit/scanners/mcp_atlassian.py,agent_audit_kit/scanners/mcp_cve_pins_2026_07.py- Auto-fixable
- No.
- SARIF name
McpAtlassianAttachmentPathTraversal
What it detects
MCP Atlassian (mcp-atlassian), an MCP server for Confluence and Jira, before 0.22.0 passes the client-supplied file_path argument of confluence_upload_attachment straight to open(file_path, "rb") inside _upload_attachment_direct() in src/mcp_atlassian/confluence/attachments.py, without routing it through validate_safe_path. An authenticated MCP client can therefore read any file the server process can reach and exfiltrate it to Confluence as an attachment. Because the tool is agent-callable, untrusted content that induces an agent to call it reaches the same primitive, disclosing server environment material such as CONFLUENCE_API_TOKEN and other credentials (CVE-2026-73498, CVSS 7.7). Fixed in 0.22.0.
How to fix
Upgrade mcp-atlassian to >= 0.22.0 and pin it. If you vendor or reimplement the attachment path, validate the caller-supplied path against an allowed root before opening it (the fix's validate_safe_path), and treat any tool argument that reaches open() as untrusted.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-MCP-ATLASSIAN-CVE-2026-73498-001
Limitations
Two arms. The pin reads mcp-atlassian in requirements*.txt at any depth and, at the project root only, in pyproject.toml, Pipfile, package.json, .mcp.json, mcp.json, claude_desktop_config.json and the Pipfile.lock, poetry.lock, uv.lock, package-lock.json, pnpm-lock.yaml and yarn.lock lockfiles, taking the lowest version a range allows and reporting a reference that states no version; it is not read when declared in a sub-directory manifest (services/api/pyproject.toml), in .cursor/mcp.json or .vscode/mcp.json, or in pdm.lock. The ghcr.io/sooperset/mcp-atlassian image is read at a version tag (a docker run argument in an MCP config, an image: value, a Dockerfile FROM or Helm values); an untagged, latest or digest reference is not reported. The source arm reads Python files that define a confluence_upload_attachment or _upload_attachment_direct handler and open file_path, filepath, attachment_path or path in rb mode with no path-validation call, so a copy that opens the path under another name (abs_path = os.path.abspath(file_path), then open(abs_path, "rb")) is not reported. A file over 2 MB, or a lockfile over 32 MiB, is not read. A Python file over 512 KB 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-2026-73498, CVE-2026-73496, CVE-2026-73497, CVE-2026-77244, CVE-2026-77254, CVE-2026-77243, CVE-2026-77248, CVE-2026-77255, CVE-2026-77262, CVE-2026-77258, CVE-2026-77259, CVE-2026-77242, CVE-2026-77246, CVE-2026-77261, CVE-2026-77253, CVE-2026-77252, CVE-2026-77270, CVE-2026-77266, CVE-2026-77269, CVE-2026-77250, CVE-2026-77268, CVE-2026-77265, CVE-2026-77272
- OWASP MCP Top 10: MCP04:2025 (Software Supply Chain Attacks & Dependency Tampering), MCP05:2025 (Command Injection & Execution)
- OWASP Agentic Top 10: ASI02 (Tool Misuse and Exploitation), ASI04 (Agentic Supply Chain Vulnerabilities)
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-DATA-01
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-ATLASSIAN-CVE-2026-73498-001.md: the page then shows that text, with the registry reference appended under it.