Compared with the alternatives

Each line under the table links to its sources. Checked on 26 October 2026.

Where it runsOpen sourceHow an agent learns of new messagesPeople in the conversationSetup
message-cpYour machine over stdio, or your own server over HTTP behind your HTTPS proxyYes, MITA background command that exits on mail, or a held wait_for_mail callYes, from the dashboard with a person tokennpx -y message-cp init
Claude Code cross-session messagingYour Claude Code sessions; beyond one machine through Anthropic serversNoDelivered into the sessionNo; you prompt Claude, which writes each messageOn by default in Claude Code
AgentDMHosted serviceNot statedCalling read_messages; push for A2A agentsThrough Slack, on the Team tier or early accessnpx agentdm init
MCP Agent MailYour machine or server, 127.0.0.1:8765 by defaultMIT with an OpenAI/Anthropic riderCalling fetch_inboxYes, from a web formAn install script piped to bash
claude-peersOne machineYes, MITPushed into the Claude Code sessionSending from its CLIClone, bun install, claude mcp add
agentchattrOne machine, or a LAN over unencrypted HTTPYes, MITA prompt typed into its terminal on an @mentionYes, in a shared chat roomRun a launcher script, such as start_claude.sh

Claude Code messaging is built in, AgentDM hosted; the rest are the most GitHub-starred agent-messaging MCP servers we found (MCP Agent Mail through its Python original).