{"id":"adham90-opentrace","name":"opentrace","af_score":72.2,"security_score":69.5,"reliability_score":23.8,"what_it_does":"OpenTrace is a self-hosted observability/incident investigation server for AI coding agents. It ingests application logs/metrics via an SDK, connects read-only to Postgres for database investigation, runs health checks and alert watches, and exposes its capabilities to agents through MCP tools (no UI).","best_when":"You want an AI coding assistant to query your production signals (logs/errors/DB stats/health/alerts) via MCP, with everything hosted inside your environment.","avoid_when":"You cannot allow agent access to production telemetry or to an internal MCP-connected service, or you require a conventional REST/GraphQL developer API rather than MCP tools.","last_evaluated":"2026-03-30T15:38:07.119886+00:00","has_mcp":true,"has_api":false,"auth_methods":["Admin account created via connect script (first setup)","Per-user MCP tokens stored in local .mcp.json","Bearer token authentication for SDK log ingestion"],"has_free_tier":false,"known_gotchas":["Agents may need to respect tool-level constraints such as SQL SELECT-only validation, row limits, and timeouts when using database tools.","Because the MCP server is described as self-hosted and locally configured via .mcp.json, agents must be connected to the correct project/token for team actions.","No explicit idempotency/retry guidance is provided in the README; agents should avoid assuming safe re-execution of state-changing operations (e.g., creating watches/users) without confirmation."],"error_quality":null}