CVE-2026-73496: MCP Atlassian: path traversal enables server file disclosure

HIGHCISA: TRACK*

Published September 14, 2026

CISO Take

MCP Atlassian, an MCP server that lets AI agents work with Jira and Confluence, has a path traversal flaw (CVE-2026-73496, CVSS 7.7) in its attachment upload tools. The server never confines a client-supplied file path, so any client with write-tool access can have the server read arbitrary local files and upload them to Atlassian. In remote HTTP, SSE or multi-user deployments that exposes server files, environment-held Atlassian credentials and potentially another tenant's data. Exploitation likelihood is modest for now: EPSS is 0.48% (top 61%), the CVE is not in CISA KEV, and there is no public exploit or Nuclei template. CISA SSVC rates it TRACK*, and the package has only 1 tracked downstream dependent, so the blast radius is limited to teams that host the server themselves. Upgrade to mcp-atlassian 0.22.0 or later, and rotate Atlassian tokens on any shared or remote instance that was exposed. The package has 38 other CVEs, so review how it is deployed overall. Local single-user stdio setups do not cross the trust boundary and are not affected.

Sources:NVDEPSSGitHub AdvisoryCISA SSVCATLAS

What is the risk?

High severity (CVSS 7.7) because the scope changes: the MCP server's filesystem and credentials become reachable by a low-privileged client over the network with no user interaction. Attack complexity is low. Real-world likelihood is moderate to low: EPSS is 0.00481 (61st percentile), there is no KEV listing, no public exploit and no scanner template, and SSVC is TRACK*. Exposure depends on deployment. Remote or multi-tenant instances are at high risk, while local stdio use is not affected. Impact is confidentiality only (C:H/I:N/A:N), but leaked Atlassian credentials could allow follow-on compromise.

How does the attack unfold?

Initial Access

A low-privileged client, or an agent steered by prompt injection, reaches the shared HTTP/SSE MCP server with write-tool access.

AML.T0053

Exploitation

The client calls an attachment upload tool with an absolute or ../ file_path that the server resolves on its own filesystem.

AML.T0037

Collection

The server uploads the targeted file, such as .env or /proc/self/environ, to an Atlassian page or issue the attacker can read.

AML.T0086

Impact

The attacker retrieves the Atlassian credentials and uses them to access other tenants' Confluence and Jira data.

AML.T0098

Initial Access

A low-privileged client, or an agent steered by prompt injection, reaches the shared HTTP/SSE MCP server with write-tool access.

AML.T0053

Exploitation

The client calls an attachment upload tool with an absolute or ../ file_path that the server resolves on its own filesystem.

AML.T0037

Collection

The server uploads the targeted file, such as .env or /proc/self/environ, to an Atlassian page or issue the attacker can read.

AML.T0086

Impact

The attacker retrieves the Atlassian credentials and uses them to access other tenants' Confluence and Jira data.

AML.T0098

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
MCP Atlassian	pip	—	No patch
2 dependents85% patched~4d to patchFull package profile →

Do you use MCP Atlassian? You're affected.

How severe is it?

CVSS 3.1

7.7 / 10

EPSS

0.5%

chance of exploitation in 30 days

Higher than 42% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Trivial

Exploitation Confidence

medium

○CISA SSVC: Public PoC

Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AVNetwork

ACLow

PRLow

UINone

SChanged

CHigh

INone

ANone

What should I do?

6 steps

Upgrade mcp-atlassian to 0.22.0 or later (fix in PR #1448, commit b041733).
Rotate Atlassian API tokens and any other secrets in the environment of remote or shared instances that ran a vulnerable version, and review Jira and Confluence audit logs for unexpected attachments.
Workarounds until patched: disable or deny the upload and attachment tools, run the server in local single-user stdio mode, or restrict network access and require strong authentication in front of HTTP/SSE endpoints.
Run the server as a non-root user in a container with a minimal filesystem and no unrelated secrets mounted.
Detection: search tool-call logs for file_path or attachments values with absolute paths or '..' segments, and look for attachments named like /etc/passwd, .env or SSH keys.
Apply least privilege to the Atlassian service account and avoid sharing one credential across tenants.

What does CISA's SSVC say?

DecisionTrack*

Exploitationpoc

AutomatableNo

Technical Impactpartial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Data ExtractionData LeakageAgentPluginAML.T0037- Data from Local System AML.T0053- AI Agent Tool Invocation AML.T0055- Unsecured Credentials AML.T0086- Exfiltration via AI Agent Tool Invocation AML.T0098- AI Agent Tool Credential Harvesting

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15-Accuracy, robustness and cybersecurity

ISO 42001

A.10.3-Suppliers

NIST AI RMF

MANAGE 2.3-Respond to previously unknown risksMAP 4.1-Risks of third-party software and components are mapped

OWASP LLM Top 10

LLM02-Sensitive Information DisclosureLLM06-Excessive Agency

Frequently Asked Questions

What is CVE-2026-73496?

Is CVE-2026-73496 actively exploited?

No confirmed active exploitation of CVE-2026-73496 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-73496?

1. Upgrade mcp-atlassian to 0.22.0 or later (fix in PR #1448, commit b041733). 2. Rotate Atlassian API tokens and any other secrets in the environment of remote or shared instances that ran a vulnerable version, and review Jira and Confluence audit logs for unexpected attachments. 3. Workarounds until patched: disable or deny the upload and attachment tools, run the server in local single-user stdio mode, or restrict network access and require strong authentication in front of HTTP/SSE endpoints. 4. Run the server as a non-root user in a container with a minimal filesystem and no unrelated secrets mounted. 5. Detection: search tool-call logs for file_path or attachments values with absolute paths or '..' segments, and look for attachments named like /etc/passwd, .env or SSH keys. 6. Apply least privilege to the Atlassian service account and avoid sharing one credential across tenants.

What systems are affected by CVE-2026-73496?

This vulnerability affects the following AI/ML architecture patterns: Agent frameworks, MCP servers, Multi-tenant AI agent deployments, Tool-calling LLM integrations.

What is the CVSS score for CVE-2026-73496?

CVE-2026-73496 has a CVSS v3.1 base score of 7.7 (HIGH). The EPSS exploitation probability is 0.48%.

What is the AI security impact?

Affected AI Architectures

Agent frameworksMCP serversMulti-tenant AI agent deploymentsTool-calling LLM integrations

MITRE ATLAS Techniques

AML.T0037Data from Local System

AML.T0053AI Agent Tool Invocation

AML.T0055Unsecured Credentials

AML.T0086Exfiltration via AI Agent Tool Invocation

AML.T0098AI Agent Tool Credential Harvesting

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.10.3

NIST AI RMF: MANAGE 2.3, MAP 4.1

OWASP LLM Top 10: LLM02, LLM06

What are the technical details?

Original Advisory

MCP Atlassian is a Model Context Protocol (MCP) server for Atlassian products (Confluence and Jira). Prior to 0.22.0, the confluence_upload_attachment and confluence_upload_attachments tools pass a client-controlled file_path through src/mcp_atlassian/confluence/attachments.py upload_attachment, and the jira_update_issue attachments parameter reaches src/mcp_atlassian/jira/attachments.py upload_attachment, without confining either path to an approved server workspace. In a remote HTTP, SSE, or multi-user deployment, absolute or traversing paths are resolved on the MCP server and uploaded to Atlassian, allowing a client with write-tool access to disclose server files, environment-held Atlassian credentials, or another tenant's data. A local single-user stdio deployment does not cross this trust boundary because the server runs in the caller's environment. This issue is fixed in version 0.22.0.

Exploitation Scenario

An organization hosts mcp-atlassian over HTTP so several teams' AI assistants share it, with a single Atlassian service token in the server environment. A low-privileged user, or an agent steered by indirect prompt injection in a Jira ticket, calls confluence_upload_attachment with file_path set to /proc/self/environ or ../../app/.env and targets a page they can read. The server reads the file from its own filesystem and attaches it to that page. The attacker downloads the attachment and obtains the Atlassian token and other secrets. They then use the token to read other tenants' Confluence and Jira data.

Weaknesses (CWE)

CWE-22Improper Limitation of a Pathname to a Restricted Directory CWE-73External Control of File Name or Path

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory: The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted directory.

[Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
[Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.

Source: MITRE CWE corpus.