Today we're launching a beta of the Skylight MCP server — a new way to bring your application's performance data directly into the AI coding assistants you already use, including Claude Code, Cursor, Zed, Claude Desktop, and Codex. Instead of context-switching between your editor and the Skylight UI to triage a slow endpoint, you can now ask your assistant about it and have it pull live data straight from Skylight.

What It Does

The Model Context Protocol (MCP) is a rapidly-emerging standard for connecting AI coding assistants to the tools and data they need to do real work. The Skylight MCP server exposes your app's performance data through MCP — endpoint latencies, aggregate traces, N+1 detections, deploy metadata, and latency trends — so your assistant can answer substantive questions about what's actually happening in production.

Ask "why is this endpoint slow?" and the assistant will pull the worst offenders, drill into the aggregate trace, and correlate the slow nodes back to specific lines in your repository. Ask "did the last deploy regress anything?" and it will pull deploy metadata, compare before-and-after latency trends, and report what changed. Because the server runs alongside your editor, it can cross-reference trace data with the code on disk — suggesting fixes, flagging N+1s in context, or recommending new instrumentation blocks.

Getting Started

Installation is a single gem install skylight-mcp, followed by a harness-specific init command that wires the server into your AI client's config. We support Claude Code directly, and emit compatible JSON and TOML snippets for every other major MCP harness. Full setup instructions, including supported platforms and example workflows, are on our MCP support page.

Beta and Feedback

MCP integration is in Beta and evolving quickly — tool shapes, arguments, and return formats will change as we learn how assistants actually use them. If you try it, we'd love to hear what's working and what isn't: which prompts produced useful answers, which tool outputs confused your assistant, and how it behaves in harnesses beyond Claude Code, which we test most heavily. Reach us at support@skylight.io.

The Future

The MCP server is the first step in a longer effort to bring Skylight's performance data directly into the tools developers are increasingly using to reason about their code. Stay tuned for future announcements!