Foxhound

Brings Foxhound's AI agent observability into your editor as an MCP server. Query production traces, compare agent runs with diff views, and inspect span timelines without leaving Claude Code or Cursor. Useful when you're debugging a flaky tool call or investigating why an agent's behavior changed between versions. The MCP layer exposes the same trace search, session replay, and regression detection you'd use in the web dashboard, but surfaces it through natural language queries in your chat window. Points at your self-hosted Foxhound instance or their cloud, authenticates with API keys, and keeps the context switching down when you're already neck-deep in agent code.