Back to Ecosystem Pulse
ProtocolExperimentalMCP
agents-can-communicate
by automatis-tools
Local-first MCP channel for agent coding sessions and handoffs
JavaScript
Updated Oct 7, 2026
Share:
Overview
Enable independent local-first AI coding sessions to exchange messages, hand off unfinished work, and coordinate parallel development across clients like Claude Code and Codex. Implements a lightweight MCP-style channel so separate agent sessions can discover peers, send reviews, and delegate subtasks without a central server. Distinctive for its local-first design and focus on developer workflows (code reviews, task handoffs, parallel edits).
Key Benefits
As agents collaborate on development tasks, knowing how they communicate and pass off work becomes essential for reproducibility and trust. This project exposes the communication layer so teams can observe handoffs, log interaction history, and reason about agent reliability and failure modes. By making inter-agent messaging explicit and local-first, it helps surface agent-to-agent evaluation signals and track agent behavior without centralized telemetry.
Ideal For
Developers prototyping multi-agent coding workflows who need peer-to-peer communication and observable handoffs between language-model-based coding assistants.
Real-World Examples
- Coordinate parallel coding by letting independent agent sessions exchange progress and merge suggestions
- Delegate unfinished tasks from one coding assistant to another and preserve handoff context
- Collect interaction logs to analyze agent failure modes and build agent track records
Works With
claude-codecodexgemini-cliopenai
Topics
agent-communicationagent-coordinationai-agentsclaude-codecodexdeveloper-toolsgemini-clilocal-firstmcpmulti-agent
Similar Tools
autogencrewai
Keywords
multi-agent trustagent-to-agent evaluationmulti-agent orchestrationmcp