ed-ai-compliance A **Model Context Protocol** server exposing the regulated-industry AI compliance knowledge from hellouchit.com as tools, resources, and prompts callable from any MCP-compatible AI client — Claude Desktop, Cursor, Zed, Windsurf, OpenAI ChatGPT, Continue, Cline, and ~40 other clients.
mcp-regulated-ai-compliance
以下摘录自该仓库的 README(片段) · 在 GitHub 读完整 README →
> A **Model Context Protocol** server exposing the regulated-industry AI compliance knowledge from hellouchit.com as tools, resources, and prompts callable from any MCP-compatible AI client — Claude Desktop, Cursor, Zed, Windsurf, OpenAI ChatGPT, Continue, Cline, and ~40 other clients.
>
> Free + open-source (Apache 2.0, dataset CC BY 4.0). Built by Uchit Vyas.
> **Disclaimer:** This is a personal, open-source, non-commercial project. Views are my own. It is not affiliated with, endorsed by, or representative of my employer.
---
Why this exists
The **OpenAI GPT Store** hosts the EU AI Act and AU AI Safety Standard coaches as ChatGPT-only assets. The **Claude Project** equivalents are private to each user's Claude Pro account (no public sharing). Neither reaches the practitioners who work primarily inside **Cursor, Zed, Continue, Cline, or the Claude API directly**.
An MCP server is the only Claude-side asset that is genuinely shareable + multi-client. It surfaces the same dataset, anti-patterns, decision trees, and classification logic — but as **tools any AI agent in any compatible client can call**. One published server → 40+ client surfaces → the practitioner who never opens ChatGPT or claude.ai still ends up citing your work.
---
What's in this folder
---
Status — v0.2.1 shipped 2026-05-29
> v0.2.1 = data-source abstraction so the same codebase runs on Node (stdio, node:http) AND on Cloudflare Workers / Deno Deploy / Vercel Edge. See worker/ for the Cloudflare scaffold.
| Phase | Status |
|---|---|
| **Phase 0 — Scope + skeleton** | ✅ done |
| **Phase 1 — Working server + reference tool** | ✅ done |
| **Phase 2 — 6 core tools** | ✅ done |
| **Phase 3 — 4 resource providers + 5 prompts** | ✅ done |
| **Phase 4 — npm publish + directory submissions** | ✅ done |
| **Phase 5 — HTTP transport + 4 playbooks + parser** | ✅ **done (v0.2.0)** |
v0.2.0 ships with
- **Streamable HTTP transport** — npx mcp-regulated-ai-compliance-http boots a Node HTTP server on port 3000 (configurable) at /mcp. Unlocks Smithery, ChatGPT MCP directory, browser-based clients, and any platform that prefers HTTP over stdio. Stateless by default; set MCP_STATEFUL=true for per-session UUIDs.
- **All 4 playbooks fully structured** — markdown parser extracts 12-week / 12-gate / phase / anti-pattern / source-URL data:
- eu-ai-act-12-weeks — Piloting → Articles 9-15 ready by 2 Aug 2026
- cisa-attestation-90-days — Federal contractor SSDF + Common Form 3201-NEW
- cloud-cost-aware-to-controlled — FinOps Aware → Controlled (AWS / Azure / GCP)
- vault-theatre-to-workload-identity — Long-lived creds → OIDC federation
- **6 tools** — lookup_control · get_anti_pattern · crosswalk · walk_playbook · classify_use_case · list_regulations
- **4 resource providers (56 URIs)** — full dataset (+ by-regulation + by-category), 15 anti-patterns (bundled + per-slug), 4 playbooks, the 20-entry crosswalk matrix
- **5 prompts** — eu-ai-act-classify · au-ai-safety-walkthrough · crosswalk-frameworks · playbook-week · anti-pattern-diagnostic