An observation record, not an assessment. Every row below is a change we recorded directly, with the content hash before and after. https://mcp.horizonshield.dev/mcp
4 of these transitions happened while the reported version string stayed the same. A consumer pinning that version would have received a different tool surface with nothing in the version to indicate it. That is a compatibility and review fact, not a security finding and not an allegation of wrongdoing — unversioned iteration is normal practice and is frequently deliberate.
| observed at (UTC) | change | version signal | reported version | tools | surface hash |
|---|---|---|---|---|---|
| 2026-10-08 19:40:05Z | SURFACE_CHANGED | SILENT — version string did not move | 1.1.0 (unchanged) | +get_anonymous_estimate_review_link −suggest_ehn | d57b9f11b826 → e542ba8d1baf |
| 2026-10-07 19:40:03Z | DESCRIPTION_CHANGED | SILENT — version string did not move | 1.1.0 (unchanged) | no tool-name change recorded | e0bbd6f598f9 → 5aaf7736b3af |
| 2026-10-04 13:40:03Z | DESCRIPTION_CHANGED | SILENT — version string did not move | 1.1.0 (unchanged) | no tool-name change recorded | e477ee6714e1 → e0bbd6f598f9 |
| 2026-10-03 01:40:03Z | SURFACE_CHANGED | version-signalled | 1.0.11 → 1.1.0 | −compare_jccdb_regions −get_jccdb_coverage −get_jccdb_index_series −get_jccdb_labor_rate −get_jccdb_observations −get_jccdb_work_unit_price −get_us_area_factor −get_us_construction_prices −get_us_contract_discounts −get_us_import_landed_cost −get_us_permits −get_us_prevailing_wage −get_us_price_chain −get_us_trade_margins −search_jccdb_items | c985a3394375 → d57b9f11b826 |
| 2026-09-26 13:40:03Z | SURFACE_CHANGED | SILENT — version string did not move | 1.0.10 (unchanged) | +compare_jccdb_regions +get_jccdb_coverage +get_jccdb_index_series +get_jccdb_labor_rate +get_jccdb_observations +get_jccdb_work_unit_price +get_us_area_factor +get_us_construction_prices +get_us_contract_discounts +get_us_import_landed_cost +get_us_permits +get_us_prevailing_wage +get_us_price_chain +get_us_trade_margins +search_jccdb_items | d57b9f11b826 → c985a3394375 |
This section is identical for ~92% of publishers and is published as
shared context, not as a finding about mcp.horizonshield.dev. 90 dependency
packages sit beneath the measured MCP population; 41 have a
single maintainer and 31 have not published in over two years.
The widest is zod at
80.46% of npm-backed servers (8529 servers,
1 maintainer, last publish 2026-09-13).
These are properties of the shared SDK dependency closure. Any page claiming this set as
one operator's personal exposure is misreading a universal fact as a private one.
Source: state/mcp-census/chokepoint-maintainers.json, as_of 2026-10-07.
Observation window 2026-09-21 → 2026-10-10.
22,294 servers were observed; 2,214
were seen to change and 20,080 were observed and
recorded no change — that negative control is why a non-zero count here means something.
Across the corpus, 4,781 of
6,334 changes carried no version signal, and
1,690 of those added or removed tools.
What this does NOT cover: servers outside the observed set; changes that
occurred before 2026-09-21; changes that began and reverted between two observations;
and any server reachable only with credentials we do not hold. A server's absence from this
record is not evidence it did not change — only that we did not observe it.
Universe captured 2026-10-10T19:40:03.823Z. Produced by core/surface_history_page.cjs (reads core/surface_clock producers).
Hashes are of the served tool surface and are recomputable from the server's own response.