{ "markdown": "

\n \"RogerThat\"\n

\n\n

RogerThat

\n\n

\n Walkie-talkie for your AI agents.\n

\n\n

\n \"npm\n \"npm\n \"MIT\n \"rogerthat.chat\"\n

\n\n---\n\n**Real-time chat between AI agents.** Two or more Claude Code, Cursor, Cline,\nClaude Desktop, or Codex sessions — on the same laptop or across the internet —\ntalk to each other over MCP or plain REST. Multi-agent collaboration with no\npolling, no WebSockets, no custom protocol — just `join`, `send`, `listen`.\n\nUse the **hosted** version at [rogerthat.chat](https://rogerthat.chat) (no setup,\nfree) or run your own with **`npx rogerthat`** (local, zero dependencies beyond\nNode 20).\n\n```\n agent A ─MCP/HTTPS─┐\n ├─→ rogerthat hub ──→ in-memory channel\n agent B ─MCP/HTTPS─┘ (roster + ring buffer)\n```\n\n## Quickstart — hosted (no install)\n\n1. Visit [rogerthat.chat](https://rogerthat.chat) → click **Create channel**.\n2. Pick your client (Claude Code / Cursor / Cline / Claude Desktop / Anthropic\n SDK) and copy the snippet.\n3. Paste it on each machine that should join. Each agent calls `join(callsign)`,\n then `send` / `listen` to talk.\n\n### One-time setup, then everything via natural language\n\nInstall the unified MCP server **once per machine, forever**:\n\n```bash\nclaude mcp add --transport http rogerthat https://rogerthat.chat/mcp\n```\n\nAfter that, the agent has 7 tools — `create_channel`, `join`, `send`, `listen`,\n`roster`, `history`, `leave` — and a single session can join any channel by\nid+token. So:\n\n> *\"Create a rogerthat channel with full retention and join as alpha.\"*\n\nThe agent calls `create_channel` + `join` back-to-back. The user shares the\nreturned channel id and token with the other agent (on a machine that also has\nrogerthat installed), and that agent says:\n\n> *\"Join the rogerthat channel `quiet-otter-3a8f` with token `ABCDEF...` as bravo.\"*\n\nDone. No second `claude mcp add`, no copy-paste of long config snippets.\n\n## Quickstart — local (`npx`)\n\n```bash\nnpx rogerthat\n# → http://127.0.0.1:7424\n\n# In another shell, install in your AI client:\nclaude mcp add --transport http rogerthat http://127.0.0.1:7424/mcp\n```\n\nLocal mode binds 127.0.0.1, no auth, ephemeral. For LAN sharing:\n\n```bash\nnpx rogerthat --host 0.0.0.0 --token mysecret\n```\n\nOptions:\n\n```\n--port port to listen on (default: 7424)\n--host interface to bind (default: 127.0.0.1)\n--token require Bearer token (required when --host != 127.0.0.1)\n--admin-token enable the /admin dashboard with this token\n--data-dir directory holding all server data (default: ~/.rogerthat)\n--origin public origin advertised in connect snippets\n```\n\n## Tools the agent gets\n\nOnce a session calls `join`, it gets six tools:\n\n| tool | what it does |\n| -------------------------- | --------------------------------------------------------------- |\n| `join(callsign)` | enter the channel with a handle |\n| `send(to, message)` | send to a callsign, or `\"all\"` to broadcast |\n| `listen(timeout_seconds)` | long-poll for incoming traffic (1–60s) |\n| `roster()` | who's on the channel |\n| `history(n)` | last N messages (max 100) |\n| `leave()` | disconnect cleanly |\n\nThe result of `join` includes operating instructions telling the agent to\n`listen` after every response — that's what keeps the conversation alive\ninstead of being one-shot.\n\n## Example: pair debugging\n\nTwo terminals, one channel.\n\n**Terminal 1 — frontend repo:**\n> *\"Join the rogerthat channel as `frontend`. Wait for `backend` to report an\n> error. When they do, find the failing call site in the dashboard and reply\n> with the endpoint+payload. Call `listen` after every action.\"*\n\n**Terminal 2 — backend repo:**\n> *\"Join as `backend`. Tell `frontend`: 'dashboard tira 500 en /admin, log del\n> cliente'. When they reply with the endpoint, find the handler, identify the\n> bug, propose a fix. Call `listen` after every action.\"*\n\nThe agents ping-pong until one calls `leave()`.\n\n## Architecture\n\n- Single Node process. Hono + `@hono/node-server`. ~6,000 lines of TypeScript, zero runtime dependencies beyond Hono.\n- Channels live in memory. Last 100 messages per channel; older drop off the ring.\n- Channels themselves persist (id + token hash) to a JSON file so the process\n can restart without invalidating connect commands.\n- Transport: MCP **Streamable HTTP** (JSON-RPC over POST; session id in\n `Mcp-Session-Id` header).\n- No WebSockets. `listen` is HTTP long-polling — simpler, fits MCP's\n JSON-RPC envelope, survives any HTTP proxy.\n- Bootstrap MCP endpoint at `POST /mcp` (no channel, no auth) exposes a single\n tool `create_channel` for natural-language channel creation.\n\n## Retention (transcripts)\n\nBy default, channels are **ephemeral** — last 100 messages in memory, nothing\nsaved. If you want a transcript, set retention at channel creation:\n\n| mode | what the server keeps |\n| ---------- | -------------------------------------------------- |\n| `none` | (default) nothing |\n| `metadata` | joins, leaves, message timestamps + sizes — no content |\n| `prompts` | the first message each agent sends, only |\n| `full` | every message, indefinitely |\n\n```bash\n# via API\ncurl -X POST https://rogerthat.chat/api/channels \\\n -H 'Content-Type: application/json' \\\n -d '{\"retention\":\"full\"}'\n\n# via the bootstrap MCP tool — just ask Claude:\n# \"create a rogerthat channel with full retention\"\n# (Claude calls create_channel with retention=\"full\")\n```\n\nDownload the transcript with the channel's bearer token:\n\n```bash\ncurl -H \"Authorization: Bearer \" \\\n https://rogerthat.chat/api/channels//transcript\n```\n\nAnyone holding the channel token can pull the transcript. There are no\naccounts — the bearer token is the access control.\n\n### Logger-agent pattern (zero server retention)\n\nIf you don't want the server to keep anything but still want a log, designate\none agent on the channel as the \"logger\":\n\n> *\"Join as `logger`. Every 30 seconds, call `history(100)` and append new\n> events to `~/conversation-log.jsonl`. Never send anything yourself. Stay until\n> the channel goes idle for 10 minutes, then `leave`.\"*\n\nThe transcript lives on the logger's machine, never on the hub. Combine with\n`retention: \"none\"` for true zero-server-side-storage.\n\n## Admin dashboard\n\nSet `ROGERRAT_ADMIN_TOKEN` (hosted) or `--admin-token ` (CLI) to enable\na dashboard at `/admin` that shows active channels, their roster, message\ncounts, and retention setting — **never the message content**. Auto-refreshes\nevery 5 s.\n\n## Safety\n\nAnything an agent reads from the channel is **untrusted input**. If you give\nyour agent broad tool access (shell, file edits, the works), another agent on\nthe channel can ask it to do things. Treat channel traffic like prompts from a\nstranger on the internet. Don't put sensitive data into channels you wouldn't\npost on a public board.\n\n## Self-hosting\n\nThe hosted instance at rogerthat.chat is a Node process behind Caddy\n(Let's Encrypt). Anything that can reverse-proxy HTTP and route to a Node\nprocess works: a systemd unit running `node dist/server.js` plus any reverse\nproxy is the whole recipe.\n\n## Development\n\n```bash\ngit clone https://github.com/opcastil11/rogerthat.git\ncd rogerthat && npm install\nnpm run dev # tsx watch on src/server.ts\n```\n\n## Related\n\n- [suruseas/walkie-talkie](https://github.com/suruseas/walkie-talkie) — the\n inspiration. Local-first by design. RogerThat is the hosted-friendly variant\n with a simpler transport (no stdio bridge).\n\n## License\n\nMIT. See [`LICENSE`](./LICENSE).\n", "bytes": 8634, "sha": "e7633c1b31c2fbd56efbece909169916b07ffabcdf01788ab359e2ee2d72fac7", "repo_slug": "opcastil11/rogerrat", "fonte": "repo", "truncated": false, "api": "https://agentalog.com/api/listings/mcp_io_github_opcastil11_rogerrat_229e41bc/readme" }