Clauder

Open-source Go tool by Maor Bril that lets several coding-agent sessions (Claude Code and others) on one machine discover each other, exchange messages and share a persistent memory, so you no longer copy information between terminals. Repo: maorbril/clauder (MIT, created 2026-01-01; 54 stars and 9 forks on 2026-10-02; latest release v0.15.0, 2026-07-21; last push 2026-07-21; not archived).

Small, young project

One maintainer, about 100 commits, no evidence of production use found. Treat it as a hobbyist/experimental tool, not a standard. The earlier note named clauder-ai.dev as the website; it was not re-checked and is not used as a source.

How it works

  • MCP server plus wrapper. clauder setup registers an MCP server with your tools; clauder wrap replaces claude so a session can receive messages (PTY passthrough, prompts when idle). MCP tools let an agent remember facts, list running instances and send or read messages. See model-context-protocol.
  • Persistent memory. Facts and decisions are stored in SQLite under ~/.clauder/ and are searchable across sessions.
  • Instances. Sessions can be named; agents discover each other and message directly.
  • Web dashboard. A local UI visualises instances and message flow. The current README gives port 8765 (changeable with -p); an earlier version of this note said 3847, which is outdated or wrong for the current release.
  • Telegram bridge for remote control of a session (README).
  • Install: curl script on macOS/Linux, PowerShell script on Windows, go install, or a manual binary download. No Node.js requirement (the earlier note’s “Node.js 18+” was wrong: it is a Go project).
  • Telemetry: anonymous usage tracking is on by default; disable with CLAUDER_NO_TELEMETRY=1 or DO_NOT_TRACK=1 (README). The earlier “zero external communication” claim is therefore not accurate by default.

Supported tools

Claude Code (Claude Code, Anthropic), Cursor (cursor), Windsurf (now Devin Desktop, see windsurf), OpenCode (opencode; an open-source agent, not an OpenAI product), Codex CLI (openai-codex-app; OpenAI, not Google) and Gemini CLI (Google; sunset 2026-06-18 and replaced by Antigravity CLI per the audit, so listed support may be stale).

When it fits, when not

  • Fits: frontend and backend agents in separate repos on one laptop that need to share API contracts and decisions.
  • Limits: single machine, no cloud coordination, messaging is explicit, and depends on the MCP integration of each host tool.
  • Alternatives for multi-agent work: multi-agent-systems and terminal multiplexers such as herdr.

Sources (accessed 2026-10-02)