Skip to content

Latest commit

 

History

History
44 lines (35 loc) · 2.69 KB

README.md

File metadata and controls

44 lines (35 loc) · 2.69 KB

First Class Agents in OpenShift (Proof of Concept)

Status

  • Proof of concept only. The repo demonstrates a potential agent contract and operator flow; it is not production ready.
  • Mock agents. All agents in mock_agent/ are synthetic services to illustrate the deployment and discovery contract we intend to enforce.
  • Security/integration gaps. Auth, RBAC, hardened reconcilers, and polished docs are still in flight.
  • The core features here overlap with Kagenti, we are investigating implementing the features demonstrated here via integration between Kagenti and OpenShift. See: Issue #12

What This Repo Demonstrates

  • Turning agent workloads into first-class Kubernetes resources via an Agent CRD and controller.
  • Defining a uniform runtime contract (/.well-known/agent.json, /healthz, /metrics) that any compliant agent image must satisfy.
  • Bridging OpenShift agents into developer tooling through an MCP server that discovers agents and relays A2A protocol messages.

Quick Start Pathways

1. OpenShift Cluster Experiment

  1. Install the Agent CRD and demo operator: cd agent-operator && oc apply -f agent-crd.yaml (plus the controller manifest once hardened).
  2. Deploy a mock agent: cd mock_agent && ./deploy.sh <AGENT_NAME> (names map to files in mock_agent/data/).
  3. Verify discovery via oc get agents and interact through the MCP bridge.
  4. Optionally, clean up with mock_agent/teardown.sh.

1. MCP Integration

  1. Install Python 3.13+ and uv or pip.
  2. Run uv sync (or pip install -e .) to pull dependencies.
  3. Start the MCP bridge: cd mcp && python oc_agent_bridge.py.
  4. Use the bridge with your IDE or CLI (e.g. claude mcp add) to list and message the mock agents.

⚠️ Treat all cluster steps as disposable demos. Hardened manifests, RBAC, and signing workflows are still future work.

Current Capabilities

  • Discover mock agents labeled with ai.openshift.io/agent.* and surface their cards.
  • Enforce the draft endpoint contract through deployment scripts and sample responses.
  • Relay user messages to mock agents via the A2A protocol using the MCP bridge.

Major Gaps Before Dev Preview

  • Security & trust foundations (TLS, signed cards, RBAC).
  • Operator hardening (stateful reconciliation, OLM packaging).
  • Conformance tooling and reusable agent base images.

Learn More

  • Architecture and positioning slides: docs/presentations/demo/.
  • Design overview and goals: docs/README.md.

Contributing & Feedback

This is an exploratory project, file issues or PRs to discuss the contract, operator shape, or feature expectations.