CVSS base score: 10.0 (Critical) vendor CNA · v3.1
Score assigned by the vendor's own CVE Numbering Authority (CNA); CISA's Vulnrichment program adds a score only when the CNA omits one. Vulnrichment source record →
Vendor: ruvnet Product: ruflo
Ruflo is an agent meta-harness for Claude Code and Codex. Prior to 3.16.3, ruflo's default docker-compose deployment exposed the MCP bridge POST /mcp and POST /mcp/:group endpoints without authentication, allowing an unauthenticated network attacker to invoke tools/call to terminal_execute, obtain a shell in the bridge container, read provider API keys, and poison AgentDB learning-store patterns. This issue is fixed in version 3.16.3.
| Date | Source | Article |
|---|---|---|
| 2026-08-03 | Check Point Research | 3rd August – Threat Intelligence Report |
| 2026-07-29 | The Hacker News | Ruflo MCP Flaw Lets Unauthenticated Attackers Run Commands and Poison AI Memory |
Articles from monitored vendor research blogs and security news feeds that specifically mention this CVE. Coverage begins when feed monitoring started; earlier articles are not indexed.
Each milestone comes from a different source, so gaps between them are gaps in what those sources publish, not evidence that nothing happened. EPSS history for this CVE holds 71 points from 2026-07-11, tracked daily from 2026-09-09; earlier points are weekly, which dates a spike to within a week. Peak probability 16.4% on 2026-09-16.
Check whether a specific version is affected →
What each source publishes about which product versions carry this vulnerability, kept as written rather than merged. Absence here means no source has published applicability data, not that nothing is affected.
Coverage across public collections only: Nuclei and OpenVAS NVT (scan);
Snort and Suricata ET Open (network); Sigma and YARA (host/SIEM);
Metasploit, Exploit-DB, public GitHub repositories and NVD exploit references (exploit). Absence
here means nothing public was found, not that detection is impossible.
GitHub PoC repositories are collected automatically by
PoC-in-GitHub (CC0) and filtered to repositories that are
still online, contain code, and are starred or at least two weeks old. They are
still not vetted: some are fakes and some are malware. Treat
them as evidence that code was published, not as code to run.
NVD exploit references are links NVD's analysts tag as exploits; they confirm
an exploit was referenced but carry no date. This product uses data from the
NVD but is not endorsed or certified by the NVD.
OpenVAS coverage derives from the Greenbone Community Feed, made available
under the Open Database License (ODbL) v1.0;
its individual Vulnerability Tests are licensed
GNU GPL v2.
| First seen | Tool | Evidence | Earliest reference |
|---|---|---|---|
| 2026-11-11 | Exploit-DB | Direct — curated exploit code | Ruflo MCP Bridge 3.16.2 - Remote Code Execution |
Metasploit is dated by the first commit of the module file, Exploit-DB by its publication date, and GitHub repositories by their creation date. GitHub PoC repositories are filtered (still online, contain code, starred or at least two weeks old) but not vetted: some are fakes and some are malware, so they show that code was published, not that it works or is safe to run. Evidence tiers are explained on the About page.
Patch and commit links and ecosystem advisory ids as published by OSV.dev. Links are listed as published, not vetted.
| If the asset is | Remediate within | Table 1 row |
|---|---|---|
| Publicly exposed | 3 days | 5 of 16 |
| Not publicly exposed | 60 days | 13 of 16 |
BOD 26-04 replaced the single KEV deadline of BOD 22-01 with four inputs: whether the asset is publicly exposed, KEV status, whether exploitation is automatable, and technical impact. Exposure belongs to each agency's asset, not to the CVE, so both outcomes are shown. Automatable and technical impact come from the CISA assessment above. Days are calendar days; for KEV entries the clock starts on the date it was added, and otherwise when the agency identifies the vulnerability on an asset. Agencies make the final exposure call asset by asset, and may set shorter timelines.
Defensive techniques inferred by MITRE D3FEND from the underlying weakness (CWE-78) — general hardening and detection guidance for this weakness class, not a vendor patch for this specific CVE.
CVE.org record → NVD entry → FortiGuard Labs → Rapid7 AttackerKB →