Skip to main content

mcp-server

Overview​

The MCP Server provides three core tools responsible for dynamically managing agents inside a Kubernetes cluster. These tools enable secure deployment, updating, and listing of agents while leveraging RBAC permissions to ensure proper governance and access control.

1. deploy_agent Tool​

Description​

The deploy_agent tool is the primary and most important tool of the MCP server. It is responsible for dynamically creating and deploying new agents inside the cluster.

Arguments​

This tool accepts three arguments:

  • name
  • description
  • skills (key-value map)

How It Works​

Using the Jinja2 templating framework, the tool dynamically generates a Kubernetes YAML manifest for the agent based on the following template:

apiVersion: kagent.dev/v1alpha2
kind: Agent
metadata:
  name: "{{ name }}"
  labels:
    {%- for key, value in skills.items() %}
    {{ key }}: "{{ value }}"
    {%- endfor %}
spec:
  serviceAccountName: "{{ name }}"
  description: "{{ description }}"
  type: BYO
  byo:
    deployment:
      image: agent/custom/template-agent:0.1.0
      env:
        - name: DB_URI
          value: "postgresql://postgres:postgres@postgres.database.svc.cluster.local:5432/ai_platform_core"
        - name: AGENT_HOST
          value: "0.0.0.0"
        - name: AGENT_PORT
          value: "8080"
        - name: AGENT_NAME
          value: "{{ name }}"
        - name: API_KEY
          value: "your-token"
        - name: PROVIDER
          value: "openai"
        - name: MODEL
          value: "gpt-4o-mini"

After generating the YAML definition, the tool uses:

  • The Kubernetes Python client
  • The RBAC permissions assigned to the MCP server within the cluster

With these permissions, the agent is automatically created and deployed.

2. update_agent Tool​

Description​

The update_agent tool is responsible for updating existing agents inside the cluster.

Arguments​

This tool also accepts three arguments:

  • name
  • description
  • skills

How It Works​

Using the provided name (which must match the existing agent), the tool updates the agent's description and skills.

The update process is performed using:

  • The Kubernetes Python client
  • The RBAC permissions granted to the MCP server

This ensures that updates are secure, controlled, and compliant with cluster governance policies.

3. list_agents Tool​

Description​

The list_agents tool retrieves information about agents currently deployed in the cluster.

How It Works​

It returns relevant metadata about existing agents.

This tool is particularly important for the main orchestrator agent, which uses the list of deployed agents to decide whether to:

  • Update an existing agent
  • Deploy a new agent

The decision is based on the current state of agents in the cluster.

Security and Access Control​

The MCP tools are exposed exclusively through AgentGateway on route /mcp-server .

All requests must pass through AgentGateway, ensuring:

  • Controlled access
  • Proper authentication and authorization
  • Secure execution of operations inside the cluster

This architecture guarantees that only authorized entities can request operations, ensuring strong security and governance.