First met in aicp-c3-session-13 · MCP & connectors
A documentation-lookup MCP: fetches current, version-specific docs for thousands of libraries so generated code uses today's APIs, not the model's memory.
Your build depends on fast-moving or obscure library versions the model may not know — otherwise modern models fetch docs themselves.
Version-specific docs for thousands of libraries, free
Embeddable inside a skill so it fires only when that skill runs (session 13's pattern)
The course's own live controversy: endorsed by one trainer, retired by another the next day
Brings current package/framework documentation into the agent's context on demand
Prevents stale-version build errors (React/Next/Node churn)
Installs per-project or inside a skill definition
Doesn't add reasoning — 'it's just a grep tool' (session 14)
Doesn't earn its tokens on well-known libraries modern models already handle
An on-demand insurance policy for deep/obscure documentation — enable for the job, then drop it.
MCP & connectors
The course itself split on it within 24 hours (s13 endorses, s14 retires); default off, enable per-job.
KB pipeline doesn't need it; relevant if SEO-o-matic builds against fast-moving frameworks.
current
claude-code: installed into Claude Code as an MCP
skills-sh: session 13 wires it INSIDE a skill so it loads only on use