Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-108583 Overview
CVE-2026-108583 is a Server-Side Request Forgery (SSRF) vulnerability in zotero-mcp versions 0.10.0 through 0.14.1. The flaw resides in the _fetch_embedded_metadata function, which retrieves URLs without validating the destination host. Attackers can leverage prompt injection to steer the Model Context Protocol (MCP) agent into invoking the zotero_add_by_url tool against loopback, private, or link-local addresses. The resulting requests, including those produced through HTTP redirects, return citation meta-tags and error details that leak information from internal services. The weakness is categorized under CWE-918.
Critical Impact
Prompt injection can coerce the zotero-mcp agent into probing internal network services, exposing metadata and error responses from otherwise unreachable endpoints.
Affected Products
- zotero-mcp version 0.10.0
- zotero-mcp versions 0.11.x through 0.14.0
- zotero-mcp version 0.14.1
Discovery Timeline
- 2026-10-10 - CVE-2026-108583 published to the National Vulnerability Database
- 2026-10-16 - Last updated in NVD database
Technical Details for CVE-2026-108583
Vulnerability Analysis
The zotero-mcp project exposes an MCP tool named zotero_add_by_url that ingests a user-supplied URL and extracts citation metadata. Internally, this path calls _fetch_embedded_metadata, which issues an outbound HTTP request to the supplied URL. The implementation does not restrict the destination to public, routable addresses. An attacker who can influence the agent's instructions, through prompt injection delivered via documents, chat context, or tool outputs, can direct the agent to call zotero_add_by_url with URLs targeting loopback (127.0.0.1), private RFC 1918 ranges, or link-local (169.254.0.0/16) hosts. The server follows HTTP redirects, so a public URL can redirect into an internal address and bypass naive URL inspection. Response data, including HTML meta-tags and error messages, is returned to the agent context, giving the attacker a signal about which internal services respond, their software versions, and configuration details. See the VulnCheck advisory for the vendor write-up.
Root Cause
The root cause is missing destination validation inside _fetch_embedded_metadata in src/zotero_mcp/tools/write.py. The function trusts the resolved address of the input URL and does not enforce an allow-list of schemes or hosts, nor does it block redirects to internal networks.
Attack Vector
Exploitation requires network access to the agent and the ability to inject a prompt that triggers zotero_add_by_url. The attacker supplies a URL, directly or via a redirector, pointing at internal infrastructure. The agent returns citation metadata or error output, which leaks information about reachable hosts and services behind the agent. User interaction is required because the agent must process the attacker-controlled content.
No verified exploitation code is published. Readers can review the HackMD technical write-up and the upstream issue tracker for proof-of-concept details.
Detection Methods for CVE-2026-108583
Indicators of Compromise
- Outbound HTTP requests from the host running zotero-mcp to loopback, RFC 1918, or 169.254.0.0/16 destinations.
- Agent transcripts or MCP tool logs containing zotero_add_by_url invocations with URLs resolving to internal addresses.
- HTTP 3xx responses from external URLs that redirect zotero-mcp toward internal hosts.
Detection Strategies
- Instrument the MCP server to log every URL passed to zotero_add_by_url and the final resolved address after redirects.
- Correlate agent tool-call telemetry with egress firewall logs to flag requests originating from the zotero-mcp process that terminate inside the trust boundary.
- Review agent conversation history for prompt-injection patterns that reference internal hostnames, cloud metadata endpoints, or private IP ranges.
Monitoring Recommendations
- Alert on any egress from the zotero-mcp process to 127.0.0.0/8, 10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16, 169.254.0.0/16, or IPv6 equivalents.
- Capture outbound DNS queries from the agent host and alert on resolution of cloud metadata hostnames such as metadata.google.internal or 169.254.169.254.
- Monitor for unusual spikes in zotero_add_by_url tool invocations tied to a single session or document source.
How to Mitigate CVE-2026-108583
Immediate Actions Required
- Upgrade zotero-mcp to a release later than 0.14.1 once the maintainer publishes a fixed version. Track the status in the upstream issue.
- Restrict the agent host's network egress to the public Zotero API endpoints required for operation.
- Treat all content ingested by the agent as untrusted and review prompt-injection controls on upstream document sources.
Patch Information
At the time of publication, no fixed version is listed in the NVD entry. Consult the GitHub repository for release notes and apply the first version that addresses _fetch_embedded_metadata destination validation.
Workarounds
- Deploy zotero-mcp behind an egress proxy that denies connections to loopback, private, and link-local address space and refuses to follow redirects into those ranges.
- Disable or wrap the zotero_add_by_url tool so that URLs are validated against an allow-list of trusted citation sources before _fetch_embedded_metadata runs.
- Run the agent in a network namespace or container with no route to internal services or cloud metadata endpoints.
# Example iptables rules to block SSRF targets from the zotero-mcp process user
iptables -A OUTPUT -m owner --uid-owner zotero -d 127.0.0.0/8 -j REJECT
iptables -A OUTPUT -m owner --uid-owner zotero -d 10.0.0.0/8 -j REJECT
iptables -A OUTPUT -m owner --uid-owner zotero -d 172.16.0.0/12 -j REJECT
iptables -A OUTPUT -m owner --uid-owner zotero -d 192.168.0.0/16 -j REJECT
iptables -A OUTPUT -m owner --uid-owner zotero -d 169.254.0.0/16 -j REJECT
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

