Triage Lunora deployment errors and migration status through MCP
Inspect Lunora logs, grouped errors, query performance and pending migrations through MCP, then return a scoped incident handoff without changing the deployment.
npx skills add agentskillexchange/skills --skill triage-lunora-deployment-errors-and-migration-status-through-mcp
Inspect an existing Lunora deployment through MCP and produce a bounded incident handoff. Invoke this workflow when an agent needs to correlate grouped errors, recent logs, query latency and pending migrations across the same deployment, rather than manually collecting each view in Lunora Studio.
What this skill actually does
Inputs and setup
Provide an owned Lunora deployment URL, an operator-approved diagnostic question, the relevant shard key if partitioned, and an observation window. Use a compatible Node.js runtime (^22.15.0 or >=24.11.0), a stdio MCP client and the exact package version below. The operator must authorize exposing operational data to the agent and its model provider: logs and grouped errors can contain user data.
Configure the MCP command as npx with arguments ["-y", "--package=@lunora/mcp@1.0.0-alpha.291", "lunora-mcp"]. Set LUNORA_URL and supply LUNORA_ADMIN_TOKEN through the client’s secret configuration. Enable LUNORA_MCP_ALLOW_OBSERVABILITY=1. Explicitly keep LUNORA_MCP_ALLOW_WRITES=0, LUNORA_MCP_ALLOW_AGENTS=0 and LUNORA_MCP_ALLOW_DATA_READS=0.
Diagnostic workflow
- Confirm the deployment and shard scope with the operator. List available MCP tools and stop if the required diagnostic tools are unavailable; do not widen permissions automatically.
- Read
lunora_get_issuesandlunora_get_logswith a small explicit limit. Correlate repeated errors and recent log evidence; keep observed facts separate from suspected causes. - Inspect
lunora_get_query_insightsfor execution counts and latency. Readlunora_get_advisoriesfor schema/query findings andlunora_get_migration_statusfor applied and pending migrations. - Return a concise handoff naming the deployment, shard, requested window, observed errors, relevant query or migration evidence, unavailable signals and proposed next checks. Redact secrets and unnecessary personal data from the report. Describe server-returned windows accurately; do not imply complete historical coverage.
Scope and limits
This workflow reads diagnostic metadata for an existing deployment. It does not initialize projects, deploy code, apply migrations, invoke application mutations/actions, start agents or traverse raw table rows. On partitioned deployments these reads are per shard. Observability tools generally accept bounded limits; migration status returns the complete migration list for its selected shard.
The deployment interface requires an admin token. Its read-only boundary is enforced by the MCP server configuration, not a restricted bearer token. Keep credentials in the client configuration and out of prompts and reports. Lunora remains alpha and API changes are possible; the setup is pinned to @lunora/mcp@1.0.0-alpha.291. Metadata and source were inspected; runtime behavior and security have not been independently audited.