OX Security examined a collection of published Model Context Protocol server listings and extracted 5,095 distinct hostnames. Its landing page reports that 2.3% of those hostnames did not resolve and that six were unregistered at survey time. A future registrant could potentially receive traffic if a client continues using one of these names. That is an…
Inspect the ClaimsCase · DiggingBeagle record
An MCP infrastructure survey found nonresolving and unregistered tool hostnames
OX Security reported 15,465 MCP server listings spanning 5,095 unique hostnames; 2.3% did not resolve, including six available unregistered domains.
Scope
OX Security survey of 15,465 published MCP server listings and 5,095 hostnames. Domain status is an observation at scan time, not a measured compromise rate.
Assessment profile
Assessment method- Setting
- evaluation
- Exploitation
- not established
- Evidence
- Grade C
- AI role
- AGAINST AI
Basis and provenance
OX reports an MCP directory sample and DNS/registration findings via an accessible landing page. The complete hostname list, scan window, sampling method, client-usage evidence and any actual domain takeover remain unavailable.
Assessed Oct 11, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-OX-MCP-DENOM · CLM-OX-MCP-DNS · OX Security: 15,465 MCP servers, 0 governance
At a glance
A hostname that was trusted when an MCP tool was installed may later be unregistered or change owners. Client pinning, server identity checks and periodic domain inventory address this continuity risk. Six unregistered names are exposure candidates, not six compromised MCP servers, and 2.3% DNS failures cannot be projected to all MCP servers.
Read the implicationsComplete list and timestamp of the nonresolving and unregistered domains. Whether any live MCP client continues to trust those names. The full report sampling method, unavailable on the landing page.
Limits and uncertaintyFull account
OX Security examined a collection of published Model Context Protocol server listings and extracted 5,095 distinct hostnames. Its landing page reports that 2.3% of those hostnames did not resolve and that six were unregistered at survey time. A future registrant could potentially receive traffic if a client continues using one of these names. That is an exposure condition, not evidence of domain takeover, credential theft or an exploit already affecting users. The publicly accessible landing page does not provide a complete sampling frame, reproducible domain list or collection dates, so the estimate must remain confined to the reported sample.
Mechanism and trust boundary
- 01
Listed tool host
An MCP registry or configuration refers to a hostname no longer under stable ownership.
Boundary: Trusted tool identifier to mutable DNS registration
- 02
Potential new registrant
A later domain owner could receive traffic if a client still connects to the old name; no exploit was shown.
Boundary: Expired domain reference to third-party endpoint
Claims & evidence
CLM-OX-MCP-DNSOX Security reports that 2.3% of surveyed hostnames failed DNS resolution and six were unregistered domains, describing them as potential takeover paths rather than confirmed exploitation.supported
Basis: reported finding
- supportsOX Security: 15,465 MCP servers, 0 governanceprimary disclosure
Key Findings, Low-Cost Domain Takeover Paths
CLM-OX-MCP-DENOMOX Security reports analyzing 15,465 published MCP server listings associated with 5,095 unique hostnames.supported
Basis: reported finding
- supportsOX Security: 15,465 MCP servers, 0 governanceprimary disclosure
Download the Report; Inside the Report
Implications
A hostname that was trusted when an MCP tool was installed may later be unregistered or change owners. Client pinning, server identity checks and periodic domain inventory address this continuity risk. Six unregistered names are exposure candidates, not six compromised MCP servers, and 2.3% DNS failures cannot be projected to all MCP servers.
Unknowns and contradictions
- Complete list and timestamp of the nonresolving and unregistered domains.
- Whether any live MCP client continues to trust those names.
- The full report sampling method, unavailable on the landing page.
Sources and citation
Material revision history
- Oct 11, 2026 · Published version · first publication · revision 123