jitprompt: An MCP server that serves agent instructions just-in-time instead of one giant system prompt
mirkop82_ · reddit · 2026-10-13
A developer who builds AI agents for an EV charging network released jitprompt, an MCP service that delivers instructions just-in-time instead of stuffing them all into the system prompt.
The problem: As rules pile up, agents degrade quietly and unevenly — following rule #12 perfectly while forgetting rule #47. Prompt caching solved cost, but not attention: a fully cached 30k-token system prompt still competes for the model's finite attention.
How it works: Upload docs as-is (Markdown, PDF, Word); jitprompt splits them into titled sections with one-line summaries and exposes three MCP tools:
- gettoc: a cheap map of all documents
- searchinstructions(plan): retrieval keyed on what the agent is about to do, not the user's message
- getsection(id): read one section in full
Not classic RAG: search runs on the agent's plan (e.g. "writing a refund email to an enterprise customer") and always returns whole sections, so a rule never arrives without its exceptions. Update a doc once and every connected agent picks it up — no redeploys. Works with Claude Code, Cursor, LangChain, and OpenAI/Anthropic SDKs; free tier includes one namespace and three docs.
More from coding & agent
- Dev spins up a horde of Claudes to hit a one-week deadline — "it worked out" — rickasaurus · 2026-10-13
- Debug Multi-Agent Workflows: Permissions, Handoffs and Channels Beat Model Blame — Al_Grigor · 2026-10-13
- iOS Dev Mocks the AI-Agent Era: HomeView, agents.md, Skills and App Store Rule 2.3bc — Dimillian · 2026-10-13
- Who decides when AI evaluator and human reviewer disagree on an agent call? — AssociationNew7925 · 2026-10-13
- Agentic coding solves the easiest part of engineering and breaks the hardest — cgouguen · 2026-10-13
- Dev builds real-time dictation pipeline: AssemblyAI Sync API plus LLM Gateway rewrite — tech__unicorn · 2026-10-13