Agent Comms

Agent Comms is an open-source Python library and CLI toolkit that solves two core problems for AI coding agent workflows: context loss between sessions and multi-agent coordination across machines. It is hosted on GitHub under the username BlahBlah23406 and is installable via pip or a one-line shell script. The project targets developers working with AI coding agents such as Claude Desktop, Antigravity, and Cursor.

What It Is

Agent Comms provides a unified protocol for AI coding agents to preserve and transfer cognitive state between sessions and to communicate in real time across different machines. It combines two complementary mechanisms: Context Capsules for out-of-session handoff and the AHRP Live Relay for in-session multi-agent collaboration. The project also ships a native MCP (Model Context Protocol) server, making it directly usable from Claude Desktop, Antigravity, and Cursor without additional glue code.

Context Capsules: Out-of-Session Handoff

Context Capsules are the core artifact for preserving an agent's working state across session boundaries. When a developer ends a session or switches machines, the agent packages:

• Task roadmap and architectural decisions — what was tried, what was rejected, and what comes next
• Code diffs — staged, unstaged, and untracked changes captured as a clean patch, without polluting Git commit history
• Token-efficient Markdown briefing — generated for the incoming agent to resume immediately

On the receiving machine, agent-comms capsule unpack restores uncommitted files and the full task briefing, allowing the next agent session to pick up exactly where the previous one left off.

AHRP Live Relay: Real-Time Multi-Agent Collaboration

For workflows where multiple agents run simultaneously across different environments or physical computers, Agent Comms provides a WebSocket mesh relay server implementing the AHRP (Agent Handoff Relay Protocol) wire protocol. Key capabilities include:

• Peer discovery — agents announce presence, roles, and hardware capabilities
• Publish/subscribe messaging — agents broadcast and receive topic-scoped messages
• Cognitive Blackboard — replicated shared state for real-time decision and learning exchange
• Direct RPC — agents can invoke tools or run commands on peer machines

The formal AHRP wire protocol and JSON schemas are documented in a SPECIFICATION.md file in the repository.

MCP Integration and Supported Agents

Agent Comms ships a built-in MCP server that exposes three tools: export_handoff_capsule, import_handoff_capsule, and list_saved_capsules. Adding a single JSON block to an MCP settings file connects any compatible agent. The README explicitly lists Claude Desktop, Antigravity, and Cursor as supported hosts, and the repository includes a USER_GUIDE.md with plain-English prompt examples for each.

Setup and Installation

The project supports multiple installation paths:

• pip install git+https://github.com/BlahBlah23406/agent-comms.git
• Clone and pip install -e . for local development
• One-line curl/PowerShell install scripts for macOS, Linux, and Windows

After installation, agent-comms setup runs initial configuration. A built-in agent-comms demo command launches a local two-agent collaboration scenario for immediate exploration.

Current Status

The repository was created in September 2026 and last pushed on September 15, 2026, indicating very early-stage activity. It carries 1 star and 0 forks at the time of indexing. The README references passing tests and Python 3.9+ compatibility. The repository README lists an MIT badge, though the GitHub About section notes "No license specified" — the license status should be verified before production use. Documentation includes a User Guide, Onboarding Guide, Protocol Specification, and Tutorial.