Agents.md

Created at 2 months ago

by benhaotang

starstarstarstarstar

Minimal MCP (Model Context Protocol) HTTP server for AGENTS.md and structured tasks, with versioned history (logs/revert) and an ephemeral scratchpad, exposed over a Streamable HTTP endpoint. The scratchpad can also be used to spawn context isolated subagents (via Gemini, OpenAI, Groq, OpenAI‑compatible, or MCP + OpenAI‑compatible) for solving focused tasks. An auth midware is also provided for user isolation and public serving. Users can also share and collaborate on projects within the server. With this context management system, the main agent retains memory across sections, but being hyper focused on the task at hand.

Categories

research-and-data

Tags

agents

mcp

context

What is Agents.md?

Agents.md is a minimal Model Context Protocol (MCP) HTTP server designed for managing AI agents and structured tasks, featuring versioned history and an ephemeral scratchpad for focused task execution.

How to use Agents.md?

To use Agents.md, you can install it via a simple script or Docker, and then create users to interact with the server. The server exposes various endpoints for managing tasks and agents.

Key features of Agents.md?

  • Hierarchical context management for AI agents
  • Scratchpads for temporary task-focused contexts
  • Subagent isolation to prevent information overload
  • Persistent knowledge across sessions
  • User authentication and collaboration capabilities

Use cases of Agents.md?

  1. Managing long-term AI projects with accumulated knowledge.
  2. Creating isolated environments for subagents to work on specific tasks.
  3. Collaborating on projects with multiple users in a shared environment.

FAQ from Agents.md?

  • Can I use Agents.md for multiple AI models?

Yes! Agents.md supports various AI models and allows for external AI integration.

  • Is there a user authentication system?

Yes! Agents.md includes an authentication middleware for user isolation and public serving.

  • How do I install Agents.md?

You can install it using a simple curl command or via Docker, as detailed in the documentation.

View More