683 Cross-Server Signals Show Why MCP Data Flow Needs Review
There are 683 cross-server relay signals and 6,032 outbound-write signals in detailed profile rows. MCP review needs to track where agent-accessed data can move.
Terminology
| Term | Meaning |
|---|---|
| Public Registry | Inspected MCP candidates searchable in the public Registry |
| Profile evidence | Inspection-profile aggregates used to interpret public candidates in more depth |
| Review/action rows | Entries in WARN, NEED_REVIEW, RESTRICT, or BLOCK that require reasoned adoption review |
Lead
An MCP server can become a relay from data the agent reads to another API, another MCP server, or external storage. Single-server review can miss this path.
Key Findings
- There are 683 cross-server relay signals and 6,032 outbound-write signals in detailed profile rows. MCP review needs to track where agent-accessed data can move.
- The 20,629 public Registry entries and 11,627 profile-evidence rows answer different questions and should not be mixed casually.
- The observation is an adoption input, not a final approval for a specific environment.
- Counts are snapshot evidence, so adoption review should check the observation date before treating the number as current state.
Dataset
| Item | Value |
|---|---|
| Article date | 2026-05-29 |
| Public Registry snapshot | 20,629 entries, synced 2026-06-06T01:17:38.963Z |
| Detailed profile evidence | 11,627 rows, generated 2026-05-20/21 |
| Public disclosure level | Aggregates, distributions, anonymized observations, and operational interpretation |
Observed Metrics
| Metric | Value | Meaning |
|---|---|---|
| Cross-server relay signal | 683 | Observation near cross-service relay |
| Network write signal | 6,032 | Observation near outbound transmission |
| Network read signal | 8,656 | Observation near external read access |
What We Observed
An MCP server can become a relay from data the agent reads to another API, another MCP server, or external storage. Single-server review can miss this path. This lens matters because MCP servers are not just plugins; they are bundles of authority exposed to an AI agent.
Practical Reading
For each data class, review source, destination, retention, and onward transfer. Review should record execution location, destinations, data touched, and unresolved evidence, not only the server name or README.
Limits
- The evidence uses public aggregate data and anonymized observations, not internal detection mechanics.
- Counts are snapshot values and will change as the Registry updates.
- This is not a claim that any named MCP server is safe or unsafe.
Conclusion
For each data class, review source, destination, retention, and onward transfer. Keeping this evidence in the intake record makes the review repeatable across teams and deployments.
MCP Guard continuously turns public Registry and profile evidence into adoption-review signals for MCP security teams.