Another week, another batch. 184 new servers push the registry from 2,638 to 2,822. Zero removed. The headline: a mature Python scraping framework with 19,000 stars enters the registry and immediately claims the top trust score. Elsewhere, game engines get dedicated API servers, a solo developer wires up 13 US government APIs, and EU compliance tooling arrives five months before the AI Act deadline.
By the Numbers
| Metric | Previous | Current | Change |
|---|---|---|---|
| Total servers | 2,638 | 2,822 | +184 |
| With source repos | 2,212 | 2,376 | +164 |
| With packages | 1,778 | 1,896 | +118 |
| Average trust score | 48.7 | 48.4 | -0.3 |
| Median trust score | 52 | 52 | — |
Trust Distribution
| Tier | Count | Change |
|---|---|---|
| High Trust (80–100) | 35 | +1 |
| Moderate Trust (60–79) | 517 | +24 |
| Low Trust (40–59) | 1,554 | +88 |
| Very Low Trust (20–39) | 712 | +71 |
| Unknown/Suspicious (0–19) | 4 | — |
The New Number One
Scrapling (io.github.D4Vinci/Scrapling) is not an MCP-native project. It is a full-featured Python web scraping framework with 19,392 stars, 1,300+ forks, and years of development behind it. Three fetcher tiers — plain HTTP with TLS fingerprint spoofing, Playwright/Chromium for dynamic pages, and a stealth mode that auto-solves Cloudflare Turnstile — plus adaptive element relocation that tracks DOM elements even when website structures change. It added MCP as a distribution channel.
The MCP integration pre-extracts and structures content before passing it to AI, reducing token consumption versus feeding raw HTML. Install with pip install "scrapling[ai]". Score: 92 — the highest trust score of any new entry we have ever recorded, and the first server to enter the top 25 at position #1. This is what it looks like when established projects decide MCP is worth supporting.
Game Engines Get API Servers
Codeturion (io.github.Codeturion) registered a matched pair of game engine API servers — unity-api-mcp (29 stars) and unreal-api-mcp (21 stars) — plus codesurface for general code analysis. The game engine servers solve a real pain point: AI coding assistants hallucinate game engine APIs constantly because documentation is vast and version-specific. These servers pre-index the full API surface — 42,000 records for Unity 6, 114,724 records across 1,019 modules for Unreal 5.7 — into local databases with sub-millisecond lookup times. The README claims 4x fewer tokens than a skilled agent and 11x fewer than a naive one during API research. Scores: 60–61.
13 Government APIs, One Developer
martc03 (io.github.martc03) registered 13 servers in a single batch — all wiring AI agents into free US government APIs. Safety recalls (NHTSA, FDA), disaster intel (FEMA, NOAA, USGS), financial data (SEC, CFPB, BLS), court records (PACER), public health (NIH), environmental compliance (EPA), grant discovery (Grants.gov), regulatory monitoring (Federal Register), and a cybersecurity vulnerability server pulling from NIST NVD, CISA KEV, and MITRE ATT&CK. All open source, all free data sources, most require no API keys. Martin, the developer, wrote on DEV.to that "government APIs are better than you think" — praising NVD 2.0's clean JSON and FEMA's geospatial query support. All score 46.
Compliance Season
Two servers arrived with regulatory deadlines in mind. eu-audit-mcp (io.github.jellewas) provides tamper-evident audit trails using HMAC-SHA256 hash chains — purpose-built for EU AI Act Article 12 (logging obligations for high-risk AI systems) and GDPR Article 30. It handles PII detection via Microsoft Presidio with European patterns, supports GDPR Article 17 erasure requests while maintaining chain integrity, and runs entirely on local SQLite. The EU AI Act's high-risk obligations take effect August 2, 2026 — five months from now. Score: 58.
Meanwhile, compliance-trestle-mcp (io.github.oscal-compass) integrates the NIST OSCAL toolchain — the standardized format for documenting security controls used in FedRAMP and government security authorization. Score: 58.
Notable New Entries
- Code Pathfinder — AI-native SAST tool that builds AST, control flow, and data flow graphs for structural vulnerability analysis. Claims 98% fewer false positives through LLM-powered triage. 110 stars. Score: 76.
- stata-mcp — Stata (the statistical software) gets an MCP bridge. 90 stars. Notable because Stata's ecosystem is notoriously closed, and its user base (economists, epidemiologists, social scientists) is far from the typical MCP developer profile. Score: 75.
- pyxel-mcp — Pyxel is a retro game engine for Python with 17,000 stars. 16-color palette, 4 audio channels, deliberate constraints mimicking classic consoles. The MCP server lets AI create pixel-art games through conversation. Score: 56.
- pump-fun-sdk — 53-tool MCP server for the Pump protocol on Solana. Full token lifecycle: creation on bonding curves, trading, liquidity management, fee sharing. 51 stars. Crypto's MCP presence continues to grow. Score: 54.
- arduino-mcp-server — AI meets hardware. Detect boards, write sketches, compile, upload, and monitor serial output. Early (3 stars) but the IoT crossover is directionally significant. Score: 59.
What to Watch
- Established projects are choosing MCP. Scrapling (19k stars) and Pyxel (17k stars) are not MCP-native — they are proven tools that added MCP as a distribution channel. This is the strongest validation signal yet that MCP is becoming a standard worth supporting, not just a playground for new projects.
- Compliance is a forcing function. EU AI Act, FedRAMP, OSCAL — regulatory deadlines create demand for tooling whether developers are enthusiastic or not. Expect more compliance-focused MCP servers as August 2026 approaches.
- Government data is underutilized. martc03's 13-server batch demonstrates that high-quality, free government APIs already exist — they just need bridges. MCP makes them discoverable by AI agents that can answer questions like "are there active NHTSA recalls on my car?" or "show me FEMA disaster declarations for my county."
- Domain expansion continues. Stata, Arduino, ParaView (scientific visualization), Pyxel — each batch brings MCP further from its developer-tooling origins into specialized professional domains.
Data sourced from the MCP Registry via MCP Scorecard. Trust scores computed from observable signals only.