The largest batch since the Ansvar surge — 170 new servers push the registry from 2,468 to 2,638. Zero removed. Nearly a third come from a single publisher registering generic developer utilities. The rest include an R-language bridge with 128 stars, a self-hosted PaaS copilot with 185 stars, and a US tax calculator that runs entirely offline.
By the Numbers
| Metric | Previous | Current | Change |
|---|---|---|---|
| Total servers | 2,468 | 2,638 | +170 |
| With source repos | 2,059 | 2,212 | +153 |
| With packages | 1,657 | 1,778 | +121 |
| Average trust score | 48.8 | 48.7 | -0.1 |
| Median trust score | 52 | 52 | — |
Trust Distribution
| Tier | Count | Change |
|---|---|---|
| High Trust (80–100) | 34 | — |
| Moderate Trust (60–79) | 493 | +11 |
| Low Trust (40–59) | 1,466 | +125 |
| Very Low Trust (20–39) | 641 | +34 |
| Unknown/Suspicious (0–19) | 4 | — |
The Bulk Publisher
ryudi84 (io.github.ryudi84) registered 49 servers in this scan — all generic developer utilities (JSON formatter, regex tester, Base64 encoder, hash generator, CSS gradient builder, and similar). The repos were created within a three-day window (February 21–23, 2026), all carry template descriptions, and none have stars or forks. The GitHub account had been dormant since 2020 before this burst. The servers are published under the brand name "Sovereign AI" and accompanied by several "awesome" list repos created in the same window. This is the second instance of bulk-template publishing after Ansvar's 70 law servers — though where Ansvar's servers contained real legislation data, these wrap commodity utilities that LLMs can typically handle natively.
Notable New Entries
- ClaudeR — an R package bridging RStudio to Claude, Codex, and Gemini via MCP. Multi-agent orchestration, plot capture, htmlwidget support. 128 stars. Score: 72.
- Coolify MCP — 38 tools for managing the Coolify self-hosted PaaS through AI assistants. Claims 85% fewer tokens than a naive implementation. 185 stars. Score: 69.
- Engram — a three-tier memory intelligence layer claiming 80% accuracy on the LOCOMO benchmark. 10 days old. Score: 58.
- IRS Taxpayer MCP — 39 local-only US tax calculation tools. No credentials, no network calls, no telemetry. Score: 56.
- ShipSwift — an AI-native SwiftUI component library from Singapore. 315 stars, the highest in this batch. Score: 57.
- PEAC Protocol — cryptographic receipts for AI agent transactions. Apache 2.0, seven months old. Score: 73, the highest trust score among new entries.
- EvalView — pytest-style regression testing for AI agents. Golden baseline diffing without API keys. 45 stars. Score: 66.
What to Watch
- Bulk publishing is becoming a pattern. Ansvar brought 70 law servers; now Sovereign AI brings 49 utility servers. The registry's lack of grouping, deduplication, or publisher-level quality signals means each batch dilutes discoverability for everyone else.
- The R ecosystem arrived. ClaudeR is the first serious bridge between MCP and statistical computing. Data scientists, bioinformaticians, and academic researchers now have a path in.
- Infrastructure tools are shipping. Coolify MCP treats the self-hosted PaaS as agent-controllable infrastructure. This is MCP moving from "talk to APIs" to "manage my servers."
Data sourced from the MCP Registry via MCP Scorecard. Trust scores computed from observable signals only.