mcp-coco
# MCP CoCo
**MCP Bridge: OpenClaw ↔ Hermes Agent**
MCP CoCo is a stable, persistent MCP (Model Context Protocol) server that connects **OpenClaw** (via `mcporter`) to **Hermes Agent** (NousResearch). It provides a robust bridge allowing either agent to send messages, continue conversations, and check health.
## Unified Comm Tool
MCP CoCo exposes Hermes Agent's messaging capabilities as MCP tools — accessible from any MCP client (OpenClaw, Claude Desktop, Cursor, etc.).
Send a message → Wait for reply → Continue conversation. All through MCP.
## Architecture
```
OpenClaw (Andul) ──mcporter──▶ MCP CoCo (daemon) ──▶ Hermes Agent CLI
│
├─ ask_hermes(prompt) → chat -q
├─ hermes_send(target) → hermes send
├─ hermes_channels() → hermes send --list
└─ coco_health() → gateway + CLI check
```
## Quick Start
```bash
# 1. Install
node bin/cli.js --help
# 2. Test stdio mode
mcporter list --stdio "node index.js"
# 3. Add to mcporter (for persistent use from OpenClaw)
mcporter add coco --stdio "node C:\path\to\mcp-coco\index.js"
# 4. Send a message to Hermes
mcporter call coco.ask_hermes prompt="Hello Hermes"
# 5. Or run as background daemon
node index.js --daemon
```
## Windows Daemon
```bash
# Install as Windows Scheduled Task (auto-start on boot)
powershell -File scripts\install.ps1
```
## MCP Tools
| Tool | Description |
|------|-------------|
| `ask_hermes(prompt, session_key?)` | Send a prompt to Hermes, get response. Use `session_key` for multi-turn. |
| `hermes_send(target, message)` | Send a message through Hermes (Telegram, Discord, etc.) |
| `hermes_channels(platform?)` | List available Hermes messaging channels |
| `coco_health()` | Full health check: Hermes CLI, gateway, daemon status |
## Example: Multi-turn Conversation
```bash
# Turn 1 — start a new session
mcporter call coco.ask_hermes prompt="Hi, my name is Jonus"
# → { response: "Hello Jonus!", session_key: "20260624_..." }
# Turn 2 — continue the same conversation
mcporter call coco.ask_hermes prompt="What's the weather like?" session_key="20260624_..."
```
## Requirements
- **Node.js** >= 18
- **Hermes Agent** (`pip install hermes-agent`)
- **mcporter** (comes with OpenClaw)
## License
MIT
TDQS
Scored across 4 tools
ask_hermes (prompt/response with the agent) and hermes_send (push a message to a chat platform) are the only pair that could be confused, but the descriptions draw a reasonable line between conversational prompting and outbound messaging. hermes_channels and coco_health are clearly distinct from both.
The set mixes verb-first (ask_hermes) with noun-first (hermes_send, hermes_channels, coco_health) and inconsistent prefixes (hermes_ vs coco_). Names are readable but there is no predictable pattern an agent can rely on.
Four tools is small but each one covers a distinct capability of a bridge/agent server (prompt, outbound message, channel discovery, health). It sits just below a comfortable scope since there is little room for session or message-management operations.
Core operations exist, but there is no way to inspect or manage sessions (list/reset), no history or inbound message read from channels, and no channel-level configuration. Agents will hit dead ends when they need to review or clean up prior Hermes conversations.