Examples of giving background agents the context to reason about a live system, so they resolve issues faster, with fewer tokens and without hallucinating.
A background agent with read access to metrics, logs, and traces, and predefined runbooks, still can't tell cause from symptom. These examples show how to add the causal context an agent needs to be effective.
Each example is self-contained and runnable.
| Example | What it shows |
|---|---|
claude-managed-agents/ |
A Claude Managed Agent investigating a live Kubernetes cluster through remote MCP servers: reaching a local cluster from a cloud agent, three auth patterns (none, static bearer, OAuth client-credentials), and evolving the system prompt from mock data to real tools. Runnable against a local kind cluster. |
open-inspect/ |
Open-Inspect's OpenCode runtime wired to a remote MCP server (Causely) so its background investigations cite root-cause and log evidence in the PRs they open, instead of guessing from source. Custom-header auth. |
aws-devops-agent/ |
AWS DevOps Agent, Amazon's managed agentic SRE service, woken by a Causely notification through a signed-webhook relay. The relay carries the diagnosis id across, so the agent calls back into Causely MCP for the causal chain instead of re-deriving one from CloudWatch. Read-only tool allowlist; OAuth client-credentials held by the service. |
- Reach. The agent runs on a different network than the cluster it's investigating and can't see
localhostor a private API server. Something bridges that gap without exposing your systems publicly. - Auth. Non-interactive, no browser. What the agent can send depends on the
client: a custom header where the client supports one (Open-Inspect), a minted
Bearertoken where it doesn't (Claude Managed Agents), or OAuth client-credentials handed to the platform at registration so nothing in the pipeline ever holds a token (AWS DevOps Agent). Same MCP server, three different auth stories.
Pick an example and follow its README:
claude-managed-agents/— a runnable end-to-end setup against a local cluster.open-inspect/— if you already run Open-Inspect and want its investigations to cite live root-cause evidence in the PRs they open.aws-devops-agent/— if you're on AWS and want Causely to wake an autonomous investigation, or an engineer to ask one.
New examples welcome, especially ones that give an agent a usable model of the system a different way, or cover a different framework or class of system. Open an issue before a large PR.