/

CVE-2026-77258

Severity:

HIGH RISK

(7.7)

Path Traversal (CWE-22) in MCP Atlassian Attachments

CVE-2026-77258

Analyzed

Vulnerability type

Category:

Path Traversal

CWE:

CWE-22

Impact:

Sensitive Information Disclosure

Severity:

HIGH RISK

(7.7)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

MCP Atlassian's Confluence attachment upload function can open a caller-selected local file without restricting access to the intended workspace. A permitted Confluence MCP caller may therefore upload readable server-local files as attachments and disclose sensitive data.

Affected context

The issue occurs in MCP Atlassian deployments where the Confluence attachment-upload capability is enabled and the server accepts a caller-controlled file path. Risk exists when the MCP server process can read files outside the intended workspace and the caller is permitted to invoke the upload operation.

Exploit preconditions

  • Attack surface: Confluence attachment upload through the MCP server
  • Required access: Permission to invoke the relevant Confluence MCP operation
  • Trigger point: Processing of the caller-supplied file_path during attachment upload
  • Why exploitation is possible: The selected path is opened without an effective restriction to an approved workspace or directory

Why this happens

The vulnerability results from trusting a caller-controlled filesystem path when opening an attachment. Without canonicalization, workspace-boundary enforcement, and an allowlisted file policy, path resolution may access files outside the intended directory. This behavior maps to CWE-22: Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal').

Potential impact

  • Sensitive server-local files may be disclosed through Confluence attachments
  • Credentials, configuration data, or application secrets could be exposed
  • Data readable by the MCP server process may be copied outside its intended location
  • Exposed information could enable follow-on compromise of connected systems

General mitigation guidance

Mitigation generally focuses on treating file paths as untrusted input and enforcing a narrow file-access boundary.

Common defensive patterns include:

  • Resolving candidate paths against a fixed workspace and rejecting paths outside its canonical boundary
  • Preferencing opaque file identifiers or controlled upload handles over raw filesystem paths
  • Applying an allowlist for permitted directories, file types, and file sizes
  • Running the MCP server with a dedicated least-privilege account and minimal filesystem access
  • Preventing access to secrets, runtime metadata, and other sensitive directories
  • Logging file-selection decisions and attachment-upload activity for review

How teams detect issues like this

Organizations can look for:

  • Static-analysis findings involving file-opening APIs that consume request or tool arguments
  • Code-review signals such as direct use of caller-provided paths, string-based path concatenation, or missing canonical-boundary checks
  • Tests that verify traversal attempts and symbolic-link escapes are rejected
  • Runtime logs showing attachment requests targeting unexpected directories or sensitive file types
  • Policy enforcement requiring workspace confinement and least-privilege filesystem permissions

Automated security review tools may help identify path-handling flows, but findings should be validated against deployment-specific access controls.

Sources

SourceLink
github.comgithub.com/sooperset/mcp-atlassian/commit/b041733473f95119dd539542a43c280737a8e460
github.comgithub.com/sooperset/mcp-atlassian/pull/1448
github.comgithub.com/sooperset/mcp-atlassian/releases/tag/v0.22.0
github.comgithub.com/sooperset/mcp-atlassian/security/advisories/GHSA-93xw-j965-9mx3

Frequently asked

questions

It can be exploitable through the MCP interface when a caller can invoke the Confluence attachment-upload operation and the server is reachable through that interface.

The attacker generally needs permission to act as a permitted Confluence MCP caller. The server process must also have read access to the targeted file.

Do not accept unrestricted filesystem paths. Resolve files only within a fixed, canonical workspace or use controlled file identifiers tied to an allowlisted storage area.

Combine strict path validation with least-privilege filesystem permissions, negative tests for traversal and symlink escapes, security-focused code review, and monitoring of file-access behavior.

Find out what's already
exploitable in your codebase.