Spikes turns UI feedback into structured data your agent can act on immediately. It exposes nine MCP tools including get_spikes for filtered feedback lists, get_hotspots to surface problem areas, submit_spike for programmatic annotations, and resolve_spike to close the loop. Reviewers click elements on live pages and rate them, capturing exact selectors and context. Your agent pulls that feedback as JSON, makes changes, and marks items resolved. Supports both local JSONL storage and a hosted backend at spikes.sh. If you're building prototypes with AI assistance and need to route human feedback back to the agent without screenshots or vague descriptions, this closes that gap.
claude mcp add --transport stdio bierlingm-spikes uvx spikes