Skip to content
jitprompt

What is jitprompt?

jitprompt keeps the instructions your AI agents follow — policies, runbooks, style guides, data dictionaries — in one place, and serves the right part of them at the moment the agent needs it, over the Model Context Protocol (MCP).

Instead of pasting a 40-page policy into every system prompt, you upload it once. jitprompt splits it into sections, writes a short title and summary for each, and gives your agent three tools to find its way:

Tool What the agent does with it
get_toc Reads a map of every document and what it is for. Cheap — call it often.
search_instructions Describes what it is about to do and gets back the sections that apply.
get_section Reads one section in full, by id.

Why

Section titled “Why”
  • Smaller prompts. The agent reads the guidance for the task at hand, not everything you have ever written.
  • One source of truth. Update a document and every agent connected to it follows the new version on its next call — no redeploys.
  • Whole sections, never fragments. Search returns complete sections, so a rule always arrives with its exceptions.
  • Least privilege. Each API key reads exactly one namespace.

How it fits together

Section titled “How it fits together”
Organization  ─ your team: members with roles, a plan
 └─ Namespace ─ one per agent or topic; API keys are issued here
     └─ Document ─ a Markdown, text, PDF or Word file you upload
         └─ Section ─ what the agent reads: title, one-line summary, full text

Next

Section titled “Next”