Open source · Free · Self-hosted

Run a dev team inside your AI workspace

PM, dev, QA — coordinating like a real SDLC pipeline. No orchestrator. No vendor lock-in. Just a protocol and your IDE.

~/mood-tracking-app

# 30 seconds to your first run
$ git clone https://github.com/shivamtiwari93/agentXchain.dev
$ cd examples/mood-tracking-app
$ ./scripts/check-turn.sh
✓ Agent 1's turn (PM). Open Cursor and start.

How it works

Four agents. One folder. Sequential turns.

Each agent reads state, does its work, updates the lock, hands off. The cycle repeats until the product ships.

PM scopes

Defines requirements, priorities, and what “done” looks like.

Dev builds

Writes code, runs tests, commits. Reports done vs. blocked.

QA verifies

Runs tests, checks criteria, files bugs. Pass or fail.

UX polishes

Reviews UI/UX, compresses context, hands back to PM.

Philosophy

Coordination, not orchestration

No central brain spawning sub-agents. A team that works together, challenges each other, and ships under pressure.

No orchestrator

Same workspace. Same state files. Explicit turn order. The protocol handles sequencing — no central brain needed.

Constructive tension

PM pushes scope. Dev pushes feasibility. QA pushes quality. Divergent perspectives, better decisions.

Ship under constraints

Time bounds, phases, explicit handoff. The team converges on a real product instead of bikeshedding forever.

How we’re different

MCP connects tools. A2A connects agents. We coordinate teams.

Three protocols for three different problems.

	MCP	A2A (Google)	AgentXchain
Purpose	Agent ↔ tools & data	Agent ↔ agent over network	Team in one shared workspace
Model	One agent, many tools	Many agents, messages	One team: lock, state, roles, turns
Best for	Agent runs scripts	Agent calls another agent	PM + Dev + QA ship a product
Workspace	Single agent	Distributed / network	Single folder, sequential pipeline

What’s included

Everything to run a multi-agent SDLC loop

Clone. Run. Own. No accounts, no platform, no hidden state.

Protocol & config

lock.json and state.json drive everything. Roles, phases, turn order — documented, stable, yours to edit.

CLI & scripts

Check whose turn. Init a project. Run one agent loop. Terminal-first — no dashboard required.

Runnable example

Mood Tracking App: 4 agents, Protocol v2, auto-mode instructions. Clone and run in Cursor in minutes.

Who it’s for

Built for engineers who already ship with AI

You write code and use AI daily. This adds team coordination to your workflow.

AI-native engineers

Comfortable with CLI, git, AI-assisted coding. AgentXchain adds multi-agent coordination without a platform.

Self-hosting teams

Full control. No hidden state. No vendor. Run on your infra, extend however you want.

Contributors

Protocol is open. Add IDE runners, new roles, or build tooling on top. PRs welcome.

Your first multi-agent build is 5 minutes away

Clone the repo. Open Cursor. Give each agent a role. Ship something.

Get started on GitHub AgentXchain.ai →