GitHub Repository

Open Source Context infrastructure for AI agents. Auto-capture and share your agents' context everywhere.

205 starsJavaScript

UltraContext – open-source shared context layer for coding agents

Name: UltraContext – open-source shared context layer for coding agents
Availability: InStock
Author: ofabioroma

by ofabioroma·Mar 8, 2026·1 point·1 comment

Visit Project View on HN

AI Analysis

●●●BangerBig BrainShip ItSolve My Problem

Real-time context sync across Claude Code, Codex, Cursor—solves actual agent isolation.

Strengths

•Auto-captures sessions across multiple agents without code changes or setup friction.
•MCP server architecture means framework-agnostic integration into any agent supporting the protocol.
•Git-like Context API provides queryable version control over agent decisions and plans.

Weaknesses

•Early adoption risk—value scales only if you actually use multiple agents daily.
•No Windows/Mac native client yet; relies on daemon + MCP, may create setup friction for non-technical users.

Post Description

Hey HN,

What Claude Code knows, Codex doesn't. What your teammate built yesterday or is building right now? The agent has no idea. Every coding agent is an island. And we as humans are the communication bottleneck between them.

UltraContext is an open source context layer that captures every agent session in realtime and makes it available to all of them. Think of it as a personal context engineer that sits between your agents so they always know what the others did.

How it works:

1. Run the daemon. It auto-captures context from Claude Code, Codex, Cursor as you work. Just work as you always do. 2. Add the MCP server to any agent. It now has real-time full awareness of what every other agent did. Ask questions, implement plans, or do whatever you want.

That's it. Two steps.

You can ask things like:

- "What did Codex change in the API yesterday?" - "Grab the plan Claude Code made and implement it" - "What's the team shipping today?" - "What did you get done this week" (Useful if you're working for X)

Consume it either via MCP or use the CLI to fork sessions and continue locally (It also supports interoperability of agents btw. Ex: Codex<>Claude Code).

We built this because we were using 3-4 coding agents daily and realized we spent more time re-explaining context than actually building. The daemon captures everything, the Context API [1] versions it like Git (create, get, append, update, delete — automatic versioning, time-travel, fork), and the MCP server distributes it to any agent that supports the protocol.

It's completely open source. Apache license.

GitHub: https://github.com/ultracontext/ultracontext Docs: https://ultracontext.ai/docs