Ramen 1.0: open-source governed MCP platform with SSO, per-tool guardrails, and eval-gated deploys
Ok_Plum3595 · reddit · 2026-11-18
A developer released Ramen 1.0, an open-source (BSD-3) self-hosted MCP platform tackling "shadow MCP" — servers multiplying on laptops with shared keys and no record. Key features:
- Tools as a repo: tools, resources, prompts, MCP Apps and skills in plain Python; deploys hit a canary first and must pass golden cases.
- No eval, no deploy: optional promptfoo suite plus NVIDIA SkillSpector scan against the canary; a group admin can halt either.
- Guardrails per tool: NeMo Guardrails (static or LLM-judged) inside the worker, before and after each call, fail-closed regardless of client.
- People, not shared keys: Entra ID / Google / OIDC sign-in, per-group roles, per-tool access, OAuth 2.1; agents get named group keys revocable in seconds.
- It directs the LLM: versioned client configs (CLAUDE.md blocks, Cursor rules) accepted per person and checked on every call.
- Two-spec MCP: 2026-07-28 and 2025-06-18 on one endpoint, Streamable HTTP and gRPC on one port, stdio bridge on PyPI.
- Every call has a name: per-call logs, audit, Arage Phoenix traces per person.
Runs on a laptop via docker compose; multi-zone and autoscaled on GKE/EKS/AKS with idle zones sleeping to zero. The 1.0 RC ran live on all three clouds (misses documented).
More from coding & agent
- Microsoft Autopilot Hits Frontier: AI Teammate Built From Zero Just 7 Months Ago — donasarkar · 2026-11-18
- Charlie Holtz shows off inline bash execution via a play button — charlieholtz · 2026-11-18
- OpenAI publishes official prompting guide for GPT-6.2 in Codex — pvncher · 2026-11-18
- AI prompt kit claims a faceless GTA 6 clipping page can make up to $10k/month — CurieuxExplorer · 2026-11-18
- Stitch launches AI agents natively in its banking OS, under bank permissions — CurieuxExplorer · 2026-11-18
- Codex 0.95 adds customizable keyboard shortcuts — charlieholtz · 2026-11-18