What does the Observatory actually measure?
The station records bounded, reproducible observations of public MCP endpoints. It does not convert reachability, schema shape or annotations into a safety, quality or vendor-trust verdict.
- Schema
- mcp-census-methodology/0.4
- Probe profile
- mcp-era-census/1.0
- Unknown = failure
- no
- Updated
- 2026-08-02
A measurement answers what was externally observable at a recorded time. It does not prove what happens behind credentials, whether tool outputs are correct, whether code is secure, or whether an operator is legally who they claim to be.
Signals measured
Whether the public endpoint answered the bounded probe at all.
Whether initialization and tool discovery produced interpretable responses for the tested protocol profile.
Whether the declared tool surface parsed and exposed the expected structural fields.
Whether names, descriptions and machine-readable annotations were present.
Whether a claim record demonstrated control of an endpoint or related record. This is not KYC.
How an endpoint's observable state changed across dated observations without rewriting prior records.
Signals explicitly not measured
Protocol Observatory does not certify security, malware absence, legal identity, business quality, correctness of tool outputs, financial reliability, suitability for a particular user or vendor trustworthiness.
A reachable server can still be harmful or incorrect. A server that requires credentials can be operational even when an anonymous probe cannot enumerate it. Unknown is preserved as unknown rather than silently scored as failure.
Protocol-era and drift handling
- Drift sample
- 2,743
- Grades changed
- 45
- Callability changed
- 24
- False-negative corrections
- 0
- Historical rows rewritten
- no
- Classifications
- legacy_only, modern_only, dual_stack, indeterminate
Method updates create new observations and explicit corrections. They do not erase the evidence lineage of what was previously measured.
Data and privacy boundary
Public outputs are limited to public-safe evidence. Raw secrets, credentials, prompts, private tool inputs or outputs, private contact details and owner-only infrastructure are outside the public Observatory surface.
The human Knowledge experience can change independently, while raw evidence URLs remain stable for external citations, badges, feeds and agents.