MCP in the wild.

Summary

The official Model Context Protocol Registry is the main public catalogue of MCP servers, and clients and downstream registries consume its metadata. We crawled its API on 23 October 2026 and analysed the server.json metadata of all 48,959 latest-version records, asking how servers are distributed, which credentials they declare, how consistent their metadata is, and how concentrated publishing is.

The study is passive and metadata-only: it reports what publishers declared, not what servers do. Its point for enterprises is what a declared secret means on a developer device: for more than a third of local servers, what the server needs is a secret stored on the machine that runs it, and a registry cannot say where those machines are.

The numbers

• 48,959latest-version server records in the official MCP RegistrySnapshot · 23 Oct 2026
• 37.4%of 18,334 local (stdio) servers declare a variable flagged as a secretRQ2 · 6,848 servers
• 93.3%of 7,171 servers that declare a credential declare no OAuth-style variable nameRQ2 · 6,688 servers
• 16.3%of all servers come from the ten largest of 30,301 publishersRQ4 · 7,989 servers

Key findings

1. Remote endpoints dominate. 65.7% of servers list one, and 96.6% of remote endpoints use Streamable HTTP.
2. Declared secrets sit on the device that runs the server. Of the 18,334 servers that ship a locally run (stdio) package, 6,848 (37.4%) declare at least one environment variable flagged as a secret.
3. OAuth-style names are rare. Registry-wide, 509 servers (1.0%) declare one. A missing OAuth-style name does not prove a static key.
4. Metadata hygiene is uneven. 11.0% of packages carry no version, and 752 credential-like variables in 421 servers are not flagged as secret.
5. Publishing is concentrated and recent. Ten publishers account for 16.3% of servers, and 59.8% of latest versions were released in September or October 2026 (release recency, not growth).

What is inside

1. 01Introduction and research questions
2. 02Background and related work
3. 03Methodology and ethics
4. 04RQ1 · Distribution and transport
5. 05RQ2 · Credential declarations
6. 06RQ3 · Metadata hygiene
7. 07RQ4 · Concentration and capabilities
8. 08Discussion: from registry to device
9. 09Limitations and conclusion
10. 10References and appendix