nimbletools-core

NimbleTools Core is a universal deployment layer for MCP tools. It enables you to turn any MCP tool—whether it is a stdio command-line utility or a native HTTP service—into a production-ready service that can be discovered, managed, and scaled within Kubernetes. The system provides a registry of ready-to-deploy MCP servers and an operator-based lifecycle that supports multi-workspace isolation, RBAC, and auto-scaling. You can deploy tools from the NimbleTools Registry, create workspaces, and then attach servers to those workspaces so they run as scalable services accessible via HTTP endpoints or stdio adapters.

To use its capabilities, start by installing the NimbleTools Core stack (via the provided install script or Helm). Once the cluster is ready, browse the MCP Registry to find a server definition (for example, an echo server or data API). Deploy a server into a workspace by creating a workspace and posting the server definition to the workspace’s servers endpoint. The deployed service becomes available through the NimbleTools API and can be consumed by any MCP client or integrated into workflows. The registry also provides examples of common tools (e.g., echo, GitHub, Postgres MCP access) to accelerate integration.

Prerequisites:

• Docker or Kubernetes cluster access (local or remote)
• curl or HTTP client
• Access to NimbleTools registry (optional, for registry-based deployments)

Step-by-step installation:

1. Install prerequisites for NimbleTools Core (as described in the project Quick Start):
   • Ensure a Kubernetes cluster is available (e.g., via k3d or a cloud provider)
   • Install Helm 3.0+ if you plan to deploy via Helm
2. Run the install script (examples from the Quick Start):

curl -sSL https://raw.githubusercontent.com/NimbleBrainInc/nimbletools-core/refs/heads/main/install.sh | bash

3. Verify deployment:

kubectl config use-context k3d-nimbletools-quickstart
kubectl config set-context --current --namespace=nimbletools-system
kubectl get pods

4. Access the NimbleTools Registry and API:

curl https://registry.nimbletools.ai/v0.1/servers | jq '.servers[].name'

5. Deploy an MCP server from the registry into a workspace (example using the echo server):

curl https://registry.nimbletools.ai/v0.1/servers/ai.nimblebrain%2Fecho > echo-server.json
# Create a workspace
curl -X POST http://api.nimbletools.dev/v1/workspaces \
  -H "Content-Type: application/json" \
  -d '{"name": "my-workspace", "description": "My first workspace"}'
# Deploy the echo server (use workspace_id from response)
curl -X POST http://api.nimbletools.dev/v1/workspaces/550e8400-e29b-41d4-a716-446655440000/servers \
  -H "Content-Type: application/json" \
  -d @echo-server.json

6. Verify deployment:

kubectl get pods -n ws-my-workspace-550e8400-e29b-41d4-a716-446655440000

Tips and common considerations:

• If you plan to deploy many MCP tools, consider provisioning a multi-tenant workspace setup with proper RBAC to isolate workloads.
• Use the NimbleTools Registry to discover pre-built servers that are tested and ready for production deployment.
• Ensure your environment variables (like NIMBLETOOLS_API_ENDPOINT) are correctly configured for your cluster and registry endpoints.
• When troubleshooting, check the NimbleTools operator and control-plane pods for status and logs, as they orchestrate the lifecycle of MCP services.
• For production, enable image signing, vulnerability scanning, and network policies to enhance security.