-
Refer to orignal repo for full instructions
-
Instructions adpated for mcpgateway kind cluster to work with an llm proxy routing to some open ai compatable backend below
docker pull quay.io/rh-ee-mmisiura/nemo-guardrails:guardrails_checks_with_tools_o1_v1 kind load docker-image quay.io/rh-ee-mmisiura/nemo-guardrails:guardrails_checks_with_tools_o1_v1 --name mcp-gateway cd plugins-adapter/plugins/examples/nemocheck/k8deploy kubectl apply -f config-tools.yaml kubectl apply -f server.yaml
-
Find url of nemo-check-server service. E.g., from svc in
server.yaml -
Update
${project_root}/resources/config/config.yaml. Add the blob below, merge if otherplugins orplugin_dirs already exists. Sample file here# plugins/config.yaml - Main plugin configuration file plugins: - name: "NemoCheckv2" kind: "plugins.examples.nemocheckinternal.plugin.NemoCheckv2" description: "Adapter for nemo check server" version: "0.1.0" hooks: ["tool_pre_invoke", "tool_post_invoke"] mode: "enforce" # enforce | permissive | disabled config: checkserver_url: "http://nemo-guardrails-service:8000/v1/guardrail/checks" # Plugin directories to scan plugin_dirs: - "plugins/examples/nemocheckinternal" # Nemo Check Server plugins
-
In
config.yamlensure keyplugins.config.checkserver_urlpoints to the correct service -
Start plugin adapter
- Open mcp-inspector to the mcp-gateway
- Try running a tool configured/not configured in nemo check config allow list in configmap E.g.