MCP tool boundaries
MCP belongs in the middle of an agent workflow: the assistant can request files, browser state, APIs, or services, while sensitive actions stay visible and permissioned.
Use this page when a private AI assistant needs controlled access to tools, files, browser context, APIs, machine-readable discovery files, and the related Clauxel modules that keep action reviewable.
Machine-readable discovery
Use llms.txt, catalog.json, agent-access.json, robots.txt, and sitemap.xml when another agent needs stable entrypoints before it follows a tool path.
Workflow context
MCP tools work best with Knowledge for private sources, Skills for repeatable procedures, Latest LLM for routing, and Console for the visible work surface.
Use MCP after the assistant knows the task boundary.
Before a tool runs, the assistant should know what source context matters, which action is allowed, what should remain human-owned, and what evidence proves the result. That is why this hub links back to Knowledge, Skills, Latest LLM, and the Console instead of standing alone.