Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
AccessLure victim to malicious web page
-
DeliveryOpen cross-origin EventSource to /sse
-
ExploitRead endpoint event, steal sessionId
-
ExecutionPOST crafted MCP tool calls to /message
-
ImpactRead/write files or execute commands on host
Vulnerability Assessment
| Weakness |
CWE
OWASP Top 10:2025
A02:2025 - Security Misconfiguration
ASVS 5.0
V13 - Configuration CORS policy restriction
|
| Exploitation Conditions | Exploitation requires the victim to be running MoLing in SSE mode, which is enabled only via the non-default -l/ListenAddr flag; stdio-mode deployments are not affected. While SSE mode is active, the victim must also visit an attacker-controlled web page (UI:R) with a browser that honors the mcp-go wildcard Access-Control-Allow-Origin: * header, which lifts the Same-Origin Policy and lets the page read the /sse stream and extract the unauthenticated sessionId. No credentials are needed by the attacker, and the advisory explicitly states that binding only to loopback (127.0.0.1) does NOT prevent the attack because it is browser-mediated from the victim's own machine. The concrete impact ceiling (local file read/write, command execution, or other actions) is bounded by which MCP tools are configured on that server. Limiting factors for defenders: the non-default SSE mode, the required user interaction, and the pre-0.5.0 version - version 0.5.0 adds mandatory bearer/query token authentication and strips the CORS header. |
| Impact | Multiple signals point to a genuinely high-priority issue, but the headline CVSS 9.6 overstates how broadly it can be triggered. The CVSS vector (AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:L) correctly reflects an unauthenticated, low-complexity, network-reachable flaw whose scope change (S:C) is justified because the browser-mediated attack reaches across trust boundaries to the host's files and command execution via configured MCP tools. The vuln.today Decision label is act_now, and the vendor GHSA advisory confirms the root cause (no auth on /sse and /message plus a hardcoded Access-Control-Allow-Origin: * from the mcp-go dependency). No EPSS score and no CISA KEV entry were provided in the input, so there is no quantitative exploitation-probability signal and no evidence of active in-the-wild use - this reads as a serious but not-yet-exploited weakness rather than a mass-exploited one. Two factors meaningfully gate real-world risk and are NOT captured by the 9.6 number: exploitation requires the victim to be running in SSE mode (the non-default -l flag) AND to visit an attacker-controlled page (UI:R), which makes this targeted/opportunistic rather than wormable. Net: treat as act_now for anyone running MoLing in SSE mode, but the absence of EPSS/KEV data and the UI:R + non-default-mode preconditions mean this is a conditional high priority, not a universally wormable critical. |
| Exploit ScenarioDerived | A victim runs MoLing in SSE mode (moling -l 127.0.0.1:6789) and, while it is running, browses to an attacker-controlled web page. Because the browser honors the wildcard CORS header, script on that page opens a cross-origin EventSource to the loopback /sse endpoint, reads the endpoint event to steal the unauthenticated sessionId, and then POSTs arbitrary MCP tool calls to /message - reading or writing local files or executing commands depending on the configured tools. The GHSA advisory includes a simplified JavaScript PoC demonstrating the sessionId extraction step. |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Oct 30, 2026
MoLing's SSE server mode (enabled with the non-default -l/ListenAddr flag) exposes the /sse and /message HTTP endpoints without authentication, and the mcp-go dependency hardcodes Access-Control-Allow-Origin: * on /sse, so any web page the victim visits can open a cross-origin EventSource, read the endpoint event, extract the unauthenticated sessionId, and then post unauthorized MCP tool calls to /message. Impact is bounded by which MCP tools are configured but can include local file read/write and command execution, and the vendor notes that binding only to loopback does not stop this because the attack is browser-mediated from the victim's own machine. Versions before 0.5.0 are affected; 0.5.0 adds mandatory token authentication and strips the wildcard CORS header. Exploitation requires the victim to run SSE mode and visit an attacker-controlled page (UI:R); no public exploit code or confirmed active exploitation is identified in the available data, though the advisory includes a simplified proof-of-concept snippet, so a public PoC reference exists but is not verified as working exploit code. The priority decision label is act_now.
RemediationAI
Vendor-released patch: 0.5.0 (release https://github.com/gojue/moling/releases/tag/v0.5.0, fix in PR https://github.com/gojue/moling/pull/48 and commit ae07f704cbc88d4419fc0fbaec278090fcd0502d); upgrading is the complete fix because 0.5.0 adds mandatory bearer/query token authentication and strips the wildcard Access-Control-Allow-Origin header, and operators should capture the auto-generated SSE auth token printed to stdout and supply it to clients as Authorization: Bearer <token> or ?token=<token> (alternatively set it explicitly with the --token/-t flag or MoLingConfig.AuthToken). Users who cannot upgrade immediately should, per the advisory, avoid using SSE mode (the -l flag) in environments where the machine may be exposed to untrusted web content, use a firewall or network policy to restrict access to the listening port while binding only to 127.0.0.1 and blocking external access, and avoid browsing untrusted websites while moling runs in SSE mode; note that loopback binding alone does not stop the browser-mediated attack, so dropping SSE mode or blocking the browsing path is the meaningful compensating control.
Affected VersionsENISA
Affected Packages
Source: OSV (Open Source Vulnerabilities)
Affects your stack
Affects 1 package in euvd
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
MoLing is a Go-based MCP (Model Context Protocol) server that provides computer-use and browser-use tooling to MCP clients. In its default stdio mode it communicates over standard input/output with the client process; when started with the -l flag it instead runs an SSE (Server-Sent Events) transport, exposing an HTTP /sse stream and a /message POST endpoint for client-to-server requests. The root cause class is CWE-942, permissive cross-domain policy with untrusted domains: the mcp-go library unconditionally emits Access-Control-Allow-Origin: * on the /sse response, which instructs browsers to lift the Same-Origin Policy for that stream, while MoLing itself performs no authentication on either /sse or /message. The SSE endpoint emits an endpoint event carrying a sessionId that is the only handle needed to talk to /message, so the wildcard CORS header effectively leaks the session credential to any origin. The upstream fix wraps the ResponseWriter with a corsRemoverResponseWriter that deletes the Access-Control-Allow-Origin header before headers and body reach the client (preserving http.Flusher for streaming), adds an sseSecurityMiddleware that requires a bearer token via the Authorization: Bearer header or a ?token= query parameter using crypto/subtle.ConstantTimeCompare, auto-generates a 32-hex-character token from crypto/rand (16 bytes) when ListenAddr is set and no token is configured, prints the token and a ready-to-use URL to stdout only, and chains sseSecurityMiddleware, requireJSONContentType (forcing a CORS preflight for cross-origin POSTs) and the SSE server. The trigger path is a browser-based cross-origin EventSource read followed by an attacker-driven POST to /message, not a memory-safety or server-side parsing flaw.
Know when Moling gets its next vulnerability
- Watch products and get an email when something new is published
- Check this CVE against your own stack and servers
- Exploit and PoC references, threat intelligence from 22 sources
No credit card · 7-day full trial
Severity by source
AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:L
Browser-mediated cross-origin read needs no privileges (PR:N) but a victim page visit (UI:R, AV:N); the HTTP server's flaw reaches host files/commands, giving S:C with C:H/I:H and conditional A:L.
Primary rating from Vendor (GitHub_M).
CVSS VectorVendor: GitHub_M
Lifecycle Timeline
2DescriptionCVE.org
MoLing is a computer-use and browser-use based MCP server. Prior to 0.5.0, when MoLing runs in SSE mode using -l, its /sse and /message HTTP endpoints do not authenticate requests and the mcp-go dependency sets Access-Control-Allow-Origin: *. A malicious web page can open a cross-origin EventSource to /sse, including a loopback listener, and read the endpoint event to obtain sessionId. After a victim running SSE mode visits that page, the attacker can use the disclosed sessionId against /message to send unauthorized MCP tool calls. Depending on the configured tools, those calls can read or write local files, execute commands or perform other actions. Binding only to loopback does not prevent this browser-mediated attack. This issue is fixed in version 0.5.0.
Technical POC Analysis
CWE-942: Permissive Cross-domain Policy with Untrusted Domains, combined with missing authentication.
Full technical POC analysis - root cause, step-by-step exploitation and patch breakdown - is part of the Pro and Business plans.
Executive ViewAI
CVE-2026-62863 is a critical vulnerability in MoLing's SSE server mode that allows any website a victim visits to hijack the MCP server and execute unauthorized tool calls, including file read/write and command execution. Leadership should care because this browser-mediated attack bypasses network protections and affects any employee running SSE mode, potentially leading to data breach or system compromise.
Apply vendor patch.
Threat Intelligence Overview
How this decision was reached
The engine reads each signal below and applies a fixed list of rules from top to bottom. The first rule that matches sets the recommended action – like a statement, every line is shown.
What every action level means
SSVC AssessmentENISA
Priority Score
Attack Technique
Information disclosure occurs when an application unintentionally exposes sensitive data that aids attackers in reconnaissance or directly compromises security.
EUVD-2026-114460
NVD
CVE.org
EPSS
GitHub
Exploit-DB