This page explains the most agent-native use case in Provenote’s current public surface.
In plain language: instead of pasting the same long research context into every coding-agent session, you can keep that context inside Provenote and let an MCP-capable host query the same notebooks, drafts, research threads, and auditable runs through one local server.
Coding agents are good at acting on context, but long copied prompts are like carrying loose papers from room to room.
Provenote’s MCP surface is closer to putting those papers into labeled folders and giving the agent a librarian desk instead of a single oversized clipboard.
That matters when you want an agent to:
Collect sources in Provenote
-> Organize them into notebooks and research threads
-> Start the first-party MCP server
-> Connect Claude Code, OpenAI Codex, Cursor, or another MCP-capable host
-> Let the host call outcome-first tools against the same local workbench
The current repo-documented MCP story is not just “there is a server.”
It is that the server is attached to outcome objects people already care about:
draft.*research_thread.*auditable_run.*That keeps the integration centered on grounded work products instead of turning MCP into a detached control plane.
The honest public claim here is:
The current repo does not claim: