Just-in-time instructions for your agents.
Upload your docs as they are. jitprompt indexes them and serves only the sections each turn needs, over MCP.
→ get_toc()
## Support Handbook [doc 3f9a1c20]
- [e812d01e] Refund Policy — 30-day window
- [7b40c9aa] Shipping Policy — rates, customs
→ search_instructions(plan="replying to a
refund request for a 6-week-old order")
# Refund Policy [id: e812d01e]
Full refund within 30 days of purchase… Store
credit may be offered instead…
Your agent reads the table of contents, not the whole manual.
Upload as-is
Drop in your Markdown, text, PDF or Word docs. jitprompt splits them into sections and writes a short summary for each, plus a one-line summary per document.
Map, then fetch
Your agent reads a compact table of contents with get_toc, opens what it needs with get_section, or calls search_instructions with a description of what it's about to do.
Whole sections, never fragments
Search returns complete sections, so a rule always arrives with its exceptions, and the model's attention stays on the guidance that applies.
Skills for any agent, loaded when the task calls for them.
Write a skill once: when to use it, the steps, the rules, an example. Your agent sees one line per skill and loads the whole thing only when a task matches, whether it runs on Claude, OpenAI, Grok, LangChain or an open-weight model. Change a skill and every agent has the new version on its next call.
Upload your SKILL.md files as they are: the description you wrote is the line your agent sees. How to write skills
## refund-emails [doc 5d1e0a7c]
Use when writing a refund email.
- [e2b8f417] Steps - [41c9d0a2] Never
## migration-review [doc a93f2b10]
Use when reviewing a database migration.
→ get_section("e2b8f417")
# Steps — confirm the order first…