Portable AI agent workspace

Carry the flow between coding agents.

my-ai-tools keeps your memory rules, MCP setup, agent commands, and reusable skills in one machine-friendly repo so you can move from one coding agent to another without rebuilding context.

Replicate the setup View the repo
Claude Code Codex OpenCode Amp Pi Cursor

$ ./cli.sh --dry-run

preview config sync across installed coding agents

$ ./generate.sh --dry-run

export local improvements back into the repo

Memory qmd, handoffs, agent notes
MCP shared server registry
Skills commands and sub-agents
One repo
Many agents
Same working context

Stop being locked inside one agent window.

Context limits, tool limits, and waiting time are real. This setup turns your AI coding workflow into portable infrastructure instead of a one-off chat history.

Jump between agents

Keep Claude Code, Codex, OpenCode, Amp, Pi, Cursor, and other assistants aligned around the same conventions, commands, and repo memory.

Bring MCP with you

Share MCP server configuration through a central registry so each agent can regain the same tools without rebuilding the setup by hand.

Keep skills close

Store reusable slash commands, sub-agent prompts, review flows, handoff templates, and project habits where every supported agent can pick them up.

A working flow for agent hopping.

The goal is simple: start with one assistant, continue with another, and keep enough context that the next agent can do useful work immediately.

1

Sync the machine

Install or preview shared configs before mutating your home directory.

2

Work in any agent

Use the same MCP registry, command skills, hooks, and repo rules.

3

Capture the handoff

Write durable learning, session notes, or explicit handoff plans before switching.

Context moves with the work

Agent A
starts the task
->
Memory
records the state
MCP
restores tools
->
Agent B
continues without waiting
handoff + memory + MCP + skills = portable coding flow

Replicate it on a machine.

Keep the repo cloneable, inspect changes before applying them, and export local improvements back to the source when your setup evolves.

Agent instructions MCP registry Command skills Review agents Memory rules Shell hooks
Clone Preview Apply 
git clone https://github.com/jellydn/my-ai-tools.git
cd my-ai-tools

./cli.sh --dry-run
./cli.sh

./generate.sh --dry-run

Use the best agent for the next step.

When a coding agent hits a limit, stalls, or lacks the right workflow, switch. Your setup should make that move cheap.

Open on GitHub Read the docs

Ask the repo.

Ask a question about the repository. The assistant answers only from the indexed README, docs, configs, and scripts, and cites the source paths.

Browser mode downloads ~9 MB index and ~300 MB model.