Story

Show HN: JitAPI – An MCP server that treats OpenAPI specs as dependency graphs

peaknk Wednesday, January 21, 2026

OP here. I got tired of writing manual wrappers for large APIs (GitHub, Stripe, etc.) just to get them working with AI agents. I found that stuffing 1,000+ endpoints into the context window usually leads to hallucinations or missing parameters.

> I built *JitAPI* to solve this by treating the API spec as a graph, not a list of tools. > *How it works:*

> 1. It ingests a raw OpenAPI spec and builds a directed graph using *NetworkX*.

> 2. It detects dependencies between endpoints (e.g., recognizing that you need an `owner_id` from `GET /user` before you can call `POST /repos`).

> 3. At runtime, it uses *ChromaDB* to find the goal tool, then traverses the graph backwards to resolve the prerequisites automatically. > > > It’s built as an *MCP (Model Context Protocol) Server*, so it works with Claude Desktop, Cursor, or any MCP client out of the box. > The stack is Python, NetworkX, ChromaDB, and PydanticAI. > Happy to answer any questions about the graph resolution logic—it's still experimental but handles the GitHub API surprisingly well.

Summary
This article provides an overview of the JIT (Just-In-Time) API, a tool for improving the performance of Java applications by dynamically optimizing code at runtime. It discusses the benefits of using the JIT API, such as faster execution and reduced memory usage, as well as its key features and how to integrate it into Java projects.
1 0
Summary
github.com
Visit article Read on Hacker News