Governed MCP for teams on GKE/EKS/AKS: SSO roles, guardrails, evals and scans gate deploys, audit.
io.github.bkraad47/ramen — Model Context Protocol (MCP) Server
A self-hosted Model Context Protocol (MCP) server for teams that deploys tools from git. It enables control over who uses the tools and supports auditing of every call. The server is identified as io.github.bkraad47/ramen and is available under the slug io-github-bkraad47-ramen.
🛠️ Key Features
Self-hosted MCP deployment for teams
Deploy tools from git
Control who uses the tools
Audit every call
🚀 Use Cases
Teams that want to run MCP tools on their own infrastructure
Deploying and managing git-based tool updates
Recording and reviewing tool invocations per call for accountability
⚡ Developer Benefits
Predictable tool provisioning via git-backed deployments
Build your MCP tools in Python, deploy them across zones on GKE, EKS and AKS.
Rust MCP node + Python 3.12 runtime workers, Streamable HTTP at the edge and gRPC inside, guardrails per tool,
Entra ID and Google Workspace sign-in, autoscaling — managed from one FastAPI console.
Streamable HTTP is the front door. Every worker serves POST /mcp — a URL and a bearer header, nothing to
install — next to the gRPC service it has had since 0.3.1, on the same port, through the same guards. The stdio
bridge stays for clients that only speak stdio.
Status. Current release 0.8.0, verified live on GKE, EKS and AKS on 2026-10-17 — smoke, the 63-check release suite
and six load-test scenarios on each cloud — then torn down; CI proves both transports on Linux and Windows. The single
record of what each run covered, and what is tested only, is
Transport → What is verified.
Build and deploy your own MCP tools across zones. Ramen turns a git repo of tools, resources and prompts
into a fleet of MCP workers behind a cloud load balancer, and keeps who-can-call-what, secrets, guardrails, canary
gates and the audit trail in one console. It is a platform for your tools, not a gateway in front of someone else's.
Each worker pairs a Rust MCP node (Streamable HTTP and gRPC on one port, bearer auth, IP allow-lists, health,
logs) with a Python 3.12 runtime that pip-installs and runs your code: user code never runs in the process that
holds the keys. The shape, the data model and why each piece is where it is:
How it works.
Git → bucket → worker. Deploy syncs the repo to a bucket; workers load by content hash. No git creds on pods.
Canary by default, gated. Roll a canary, smoke-test tools/list, diff its tool schemas against stable
(breaking changes stop unless you say breaking: true), run the repo's golden cases, then roll stable. Failure
leaves stable untouched.
Multi-zone from day one. Group → Environment → Zone → Worker; the load balancer routes on ramen-group /
ramen-zone, so one client config works for every zone.
Guardrails inside the worker. Opt a tool into a check before the call and after the result, run in the Python
runtime next to it: NVIDIA NeMo Guardrails rails (deterministic or
LLM-backed, your model and key) or a policy.py of your own. A blocked call is a readable tool error and the access
log names the rail. Fail-closed by default.
Your identity provider decides, your admins override. Add Microsoft Entra ID, Google Workspace or any OIDC
provider on the Config page, on any cloud; read a person's groups from Graph or Cloud Identity at sign-in, map them
to roles per group, and keep what admins set by hand.
Roles per group, and roles you define. Group Admin, Viewer, MCP User or a custom role on one of those bases
(analyst on Viewer), usable in memberships, provider rules and per-tool access; a role can be OAuth only.
Decide per environment who may see and who may call each tool; a drift signal flags a caller repeating one call.
Autoscaling by Ramen. Each zone runs between 1 worker and a max that admins set; the console scales up fast
under load and back down when quiet, the same on all three clouds, with no HPA and every step in the audit log.
Three clouds. Terraform and Helm for GKE, EKS and AKS, with the same console and guards on each
(Deploy to the cloud).
Enterprise controls.rmk_ MCP keys and rmn_ API keys stored as salted hashes, IP rules per zone at the node
and at the cloud edge, secrets that are never displayed, Redis throttles, an audit line for every action.
One Deployment per zone, not one per tool. A worker loads every tool of the group, so thirty tools run on
one Deployment per zone (two pods with a canary) behind one load balancer; autoscaling adds pods for load, not for
tool count.
Needs Docker with compose v2, uv, git and make. make up pulls the release's two images from Docker Hub,
about a minute; make up BUILD=1 builds them from source instead (three to five minutes).
sh
git clone https://github.com/bkraad47/ramen && cd ramen
make up # Firestore emulator + console https://localhost:8443 + one worker (localhost:8080: Streamable HTTP + gRPC)
make demo # zone, group `demo` from the demo repo, an rmk_ key, a canary deploy, then a tools/call# PASS: demo_calculator_tool(2,3,add) -> 5 <- the success line
open https://localhost:8443 # self-signed cert; admin@ramen.local + the generated password: grep ADMIN_PASSWORD deploy/local/.env
make down # stop and remove volumes
make demo is safe to re-run: an existing zone, group or environment answers "exists" and a fresh key is
generated each time. Nothing in the stack is a static secret: make up generates the admin password, Fernet key,
MCP key, admin key and session secret into deploy/local/.env (git-ignored), and scripts/check_secrets.py fails CI
if a literal one ever returns.
Connect a client
A worker is a URL (/mcp) and two routing headers, ramen-group and ramen-zone. The credential decides whose
name a call runs under:
A group key (rmk_) in Authorization: Bearer, for agents and automation. Any HTTP-capable client.
A person's sign-in through the console (OAuth 2.1, with a Ramen password, Entra ID or Google): Claude Code
over HTTP, or any stdio client through the bridge.
Identity providers: Entra ID with group lookup and role rules
Custom roles, one OAuth-only
Zone actions: count, max, last autoscale decision
Zones and workers with the max per group
Transport and security, in short
Workers speak Streamable HTTP (POST /mcp, MCP spec 2025-06-18) and gRPC (ramen.v1.Mcp/Call) on one port
through one implementation of every guard, so a check is on both transports or on neither. TLS ends at the load
balancer. The node checks the credential, the source range at the provider's x-forwarded-for hop, blocked names
and the size and in-flight caps. The runtime reaches the bucket as the zone's own cloud identity, and guardrails run
between the runtime and your tool. Hop by hop, with what protects each and what is verified:
Transport. Boundaries and non-goals:
Threat model.
Deploy to the cloud
Target
What Terraform and Helm build
Guide
GCP
GKE Autopilot, Firestore, GCS, Secret Manager, global HTTPS LB (GKE Gateway, header-routed gRPC and Streamable HTTP), Cloud Armor
Bring-up is the same on each: terraform apply → helm upgrade --install (the release's public images, or
make push to your own registry) → add a zone and a group in the console → Deploy. About 25 minutes, mostly waiting
for the cluster and the load balancer. The edge gets a publicly trusted certificate on a free sslip.io hostname
(Let's Encrypt on AKS), so clients use https://<public_hostname>/mcp and the bridge points at
<public_hostname>:443 --tls with nothing to import. What each cloud run covered:
Transport → What is verified.
Write your own tools
A group repo is any git repo with mcp/tools/<name>/<name>.py + <name>.json (and resources/, prompts/,
requirements.txt). Secrets are referenced as {{$group.NAME}} or exported through mcp/env.yaml; nothing about
the transport leaks into tool code. Declare output and the worker publishes it as outputSchema and validates every
result; add mcp/tests.yaml and your cases gate every deploy; add mcp/guardrails.yaml and NeMo rails or a
policy.py check the arguments before a tool runs and its result before it leaves the worker. Start from
ramen-demo-mcp-group; the contract, local development and heavy
dependencies: The MCP repo ·
Guardrails.
make up BUILD=1 # the compose stack from source instead of the published images
make test# runtime-py and console (pytest, 90% coverage gate) plus node-rs (fmt, clippy, test)
make test-harness # tests/: conformance + e2e (skips without a running stack)
make proto # regenerate Python stubs from proto/ (Rust stubs build via tonic-build)
make demo-worker # node + runtime locally without Docker
uv run --project docs --group docs mkdocs serve # docs at http://127.0.0.1:8000
Contributing
See CONTRIBUTING.md — use it, fork it, change it, with attribution; renaming it as a new commercial product of your own is not acceptable.