Features

Everything the loop needs — Build, Use, Measure.

One object — the skill — on one surface — your MCP endpoint. Author it, serve it to every agent, and see exactly what got used.

Build

An authoring studio, not a text box.

Discoverability is the whole game. The studio keeps your description and trigger phrases sharp — because they decide whether a skill is ever found.

Authoring studio

Write multi-file skills — SKILL.md plus supporting docs, scripts, and templates — in one editor.

Draft with AI

Generate a strong first pass at the name, description, and trigger phrases, then refine — not a black box.

Discoverability-first

Inline nudges keep descriptions distinct and trigger phrases well-scoped — what actually gets matched.

Multi-file skills

Attach instructions, docs, scripts, and templates with typed kinds. Agents fetch each only when needed.

Immutable versioning

Every save creates a new version. Old versions are never mutated, so history is always reproducible.

Org-scoped by default

Skills, files, events, and billing belong to an organization. A personal workspace is created on signup.

Use

Connect once. Every agent reads at runtime.

A hosted MCP server your agents talk to directly. Progressive disclosure keeps their context small and your costs down.

One hosted endpoint

Point your agent at a single MCP URL once. No proxies, no wrappers, no per-skill plumbing.

Works with any MCP client

Claude Code, Cursor, Copilot, Windsurf, LangChain/LangGraph, OpenAI Agents, Vercel AI, Mastra.

Progressive disclosure

find_skills → get_skill → get_skill_file. Agents pull only the detail they need, keeping context lean.

API keys & OAuth

Org-owned API keys for headless callers, plus an OAuth provider for agent clients that need it.

Measure

Three truths nobody else can show you.

The query is the telemetry: usage is captured as a byproduct of serving each request. Every insight links straight back into the editor.

Used vs. dead skills

See what's pulling its weight and what you can safely deprecate.

Discovery gaps

Searches that matched nothing good — the prioritized list of skills to write next.

Surfaced, not loaded

Skills that show up in results but never get loaded — a description or scoping problem.

Dead reference files

Bundled files agents never fetch — trim them to keep skills tight.

Spotlight · Discovery gaps

See what your agents looked for and never found.

Every search is logged with its best similarity score. When nothing matched well, that's a gap — a skill you should write, or a description you should fix. Each one links straight into the editor.

Start measuring
mock api responses for testsno match
write terraform for gcp0.38
generate an openapi spec0.44