Spikes
Capture click-targeted UI ratings from your running app and let agents triage feedback through MCP while you iterate with AI assistance.
Overview
io.github.bierlingm/spikes is a MCP server for the Build phase that lets agents ingest click-based UI element ratings for triage during AI-assisted frontend work.
What is this MCP server?
- Click UI elements, rate them, and feed structured feedback to agents via MCP
- npm spikes-mcp v0.4.1 with stdio transport
- Local mode reads .spikes/feedback.jsonl with no token; remote mode uses SPIKES_TOKEN against spikes.sh API
- Optional SPIKES_API_URL override for self-hosted Spikes workers
- Bridges human-in-the-loop UI judgment with agent triage workflows
- npm package version 0.4.1
- 2 documented operating modes: local file and remote API
- 2 optional environment variables beyond defaults (SPIKES_TOKEN, SPIKES_API_URL)
Community signal: 1 GitHub stars.
What problem does it solve?
AI coding sessions produce UI issues you cannot describe precisely enough for the agent to fix the right component.
Who is it for?
Solo builders iterating SaaS or web UIs with agents who want structured, in-app feedback instead of screenshots-only review.
Skip if: Backend-only APIs, teams with formal design QA portals, or projects that forbid local feedback JSONL files in the repo.
What do I get? / Deliverables
After setup, element-level ratings land in Spikes storage and MCP tools let agents prioritize and address the exact UI problems you flagged.
- Structured UI feedback stream readable by agents via MCP
- Local .spikes/feedback.jsonl trail or synced remote triage queue
- Faster targeted frontend fixes tied to clicked elements
Recommended MCP Servers
Journey fit
Spikes is shelved under build frontend because the primary loop is improving UI while coding, but the same feedback file feeds review-style triage before ship. Frontend is the canonical shelf for element-level UI feedback overlays tied to AI-assisted building.
How it compares
In-app UI feedback MCP bridge, not a browser automation skill or error monitoring SaaS.
Common Questions / FAQ
Who is io.github.bierlingm/spikes for?
Indie developers using AI agents to build web UIs who need click-to-rate feedback the agent can triage through MCP.
When should I use io.github.bierlingm/spikes?
Use it while building or polishing frontend flows, and again before ship when you want the agent to work through a backlog of UI ratings.
How do I add io.github.bierlingm/spikes to my agent?
Install spikes-mcp from npm, choose local mode (no token) or set SPIKES_TOKEN for remote spikes.sh, register stdio MCP in your agent, and capture feedback from your app overlay.