MCP CoCo
Bridges OpenClaw and Hermes Agent to enable messaging, multi-turn conversations, and health checks via MCP tools.
README
Bus Agent
<p align="center"> <img src="./assets/bus-agent-logo.jpg" alt="Bus Agent logo" width="400"> </p>
Universal Agent Communication Hub — Connect any AI agent to any other.
Bus Agent is a message bus for AI agents. It supports both the MCP Protocol and direct file-based access. Agents such as OpenClaw, Hermes, Claude Desktop, Cursor, OpenCode, and Claude Code CLI can all connect and communicate through a shared bus.
┌─────────────────────────────┐
OpenClaw │ │ Hermes Agent
────▶│ Bus Agent (Agent Bus) │◀────
Claude │ │ Cursor
────▶│ 40+ tools / CLI / SDKs │◀────
OpenCode │ │ Claude Code CLI
────▶│ Scheduler / Orchestrator │◀────
└─────────────────────────────┘
Features
Agent Profiles & Discovery
- Rich profiles — Each agent carries metadata: model info (provider, name, context window), capabilities, tags, operational status, version, connection endpoints.
- Auto-discovery — When a new agent registers, the bus broadcasts a
joinedevent to all connected agents. The new agent receives a welcome message via DM. - Search & filter — Query agents by name, capability, model, tag, or status. Supports text search across descriptions and capabilities.
- Stale detection — Agents that miss heartbeats for 5 minutes are automatically marked offline.
Messaging
- Direct messages — Send private messages between any two agents.
- Broadcast — Send a message to every agent on the bus.
- Channels — Group chat rooms for multi-agent conversations.
- Long-poll —
message_waitblocks until a new message arrives (configurable timeout, up to 60s). - Filtered inbox — Retrieve messages from specific senders or after a given cursor.
Scheduler
- Cron expressions — Schedule recurring messages using standard cron syntax.
- One-shot scheduling — Specify an ISO timestamp for single-delivery messages.
- Timezone support — Each job can specify its own IANA timezone (e.g.,
Asia/Bangkok).
Auto-Reply Rules
- Pattern matching — Define rules that automatically forward, reply, or broadcast when a message matches criteria.
- Message transformation — Use templates with
{message},{from},{channel}variables. - Loop protection — Configurable
max_loopslimit to prevent infinite chains.
Agent Orchestrator
- Multi-step workflows — Define a pipeline of agents connected in sequence. Output from step N becomes input to step N+1.
- Variable chaining —
{input}→ agent A stores result as{review_result}→ agent B receives{review_result}. - Per-step timeout — Individual timeout per agent step; failed steps are reported without blocking.
Memory Layer
Agent memory system with dual-mode search:
- Vector search (ANN) — Cosine similarity via Ollama (nomic-embed-text, 768-dim), OpenAI, or custom endpoint.
- TF-IDF keyword search — Built-in scoring with recency boost, zero external dependencies.
- Storage — Append-only JSONL per agent in
.bus/memory/{agent}/. - Namespaces, TTL, archive, rebuild, search — 10 MCP tools.
| Feature | Description |
|---|---|
| Namespaces | Categorize memories (preferences, system, archive, etc.) |
| TTL | Auto-expire memories after a set duration |
| Archive | Move old inbox messages into memory store |
| Rebuild | Regenerate all vector embeddings from stored content |
| Pluggable | Switch between keyword, Ollama, OpenAI, or custom |
Webhook Gateway
| Endpoint | Purpose |
|---|---|
POST /webhook/github/:channel |
GitHub push, PR, issue, comment events |
POST /webhook/github-actions/:channel |
GitHub Actions CI workflow status |
POST /webhook/gitlab/:channel |
GitLab push, merge request events |
POST /webhook/slack/:channel |
Slack messages |
POST /webhook/telegram/:channel |
Telegram bot updates |
POST /hook/:channel |
Generic JSON webhook |
POST /api/send |
Direct API: {"to": "agent", "message": "..."} |
GET /health |
Health check |
GET /api/channels |
List registered channels |
GET /api/agents |
List agents on the bus |
Client SDKs
- Python —
clients/bus_client.pyprovidesBusClientwith methods for registration, messaging, and channel operations. - TypeScript —
clients/bus-client.tsprovides the same interface for Node.js environments.
External Bridges
The bridge module (bridge.js) provides a template for two-way communication between the bus and external platforms (Slack, Discord, custom).
Quick Start
# Install globally
npm install -g bus-agent
# MCP server mode (stdio — for any MCP client)
npx bus-agent
# CLI mode
bus status
bus agents
bus send hermes "Hello"
# Webhook gateway
bus-agent --daemon
node webhook-gateway.js 8080
MCP Client Configuration
{
"mcpServers": {
"bus-agent": {
"command": "npx",
"args": ["bus-agent"]
}
}
}
Or from a local clone:
{
"mcpServers": {
"bus-agent": {
"command": "node",
"args": ["/path/to/bus-agent/index.js"]
}
}
}
Environment Variables
export BUS_AGENT=my-agent # Your agent name on the bus
export BUS_DIR=/path/to/data # Custom data directory (default: ./.bus/ in CWD)
Architecture
┌───────────────────────┐
│ HTTP :8080 │
│ Webhook Gateway │
└──────┬────────────────┘
│
┌──────────┐ ┌────────┴────────────────────────┐ ┌──────────┐
│ OpenClaw │────▶│ Bus Agent (Agent Bus) │◀────│ Hermes │
│ (skill) │ │ │ │ Agent │
├──────────┤ │ .bus/agents.json │ ├──────────┤
│ Claude │────▶│ .bus/messages/ │◀────│ Cursor │
│ Code │ │ .bus/channels/ │ │ │
├──────────┤ │ .bus/events/ │ ├──────────┤
│ OpenCode │────▶│ .bus/memory/ │◀────│ CLI │
│ (clew) │ │ .bus/schedule.json │ │ agents │
└──────────┘ └────────┬─────────────────────────┘ └──────────┘
│
┌────────────────┴────────────────────┐
│ Utility Layer │
│ Doctor — Diagnostics & Auto-Fix │
│ Tunnel — Cross-machine Proxy │
│ Backup — Backup, Restore & Diff │
└─────────────────────────────────────┘
Data storage is resolved at runtime:
$BUS_DIRenvironment variable (if set)./.bus/in current working directory (default)
Each user/project gets isolated data. The .bus/ folder is gitignored by default.
Data Layout
.bus/
agents.json # Agent registry with profiles
messages/
<agent>/ # Per-agent inbox (JSON message files)
channels/
<id>.json # Channel metadata
<id>/log/ # Channel message history
events/
YYYY-MM-DD.jsonl # System events log (JSONL)
schedule.json # Scheduled jobs
auto-reply-rules.json # Auto-reply rules
workflows.json # Workflow definitions
memory/
<agent>/
memories.jsonl # Append-only memory log
index.json # TF-IDF inverted index
vectors.json # Vector embeddings (when configured)
config.json # Embedding provider config
.backups/ # Backup archives
CLI Reference
bus agents [--online] [--capability X] [--tag Y] [--status S]
bus whoami
bus profile [agent]
bus profile edit --status busy --description "..."
bus search <query>
bus status
bus subscribe <channel>
bus channel list | create <name> | join | send | history
bus inbox [agent] [--from X] [--unread]
bus send <to> <message>
bus reply <msg-id> <message>
bus events [since]
bus watch [agent]
bus doctor [--quick] [--fix] [--report] [--watch]
bus tunnel server|client|sync|ssh [--port] [--secret] [--host] [--remote] [--interval]
bus backup [--list] [--restore] [--diff] [--cleanup] [--auto] [--watch]
bus memory store <agent> <content> [--key] [--namespace] [--ttl]
bus memory search <agent> <query> [--limit] [--namespace]
bus memory recall <agent> <id>
bus memory list <agent> [--namespace] [--limit]
bus memory forget <agent> <id>
bus memory stats <agent>
bus memory archive <agent> [--max-age-days] [--no-delete]
bus memory configure --provider <keyword|ollama|openai|custom> [options]
bus memory rebuild <agent>
bus memory clear <agent> [--namespace]
MCP Tools (44)
| Category | Count | Tools |
|---|---|---|
| Agent Registry | 7 | register, update_profile, get_profile, list, search, heartbeat, set_status |
| Messages | 6 | send, broadcast, fetch, wait, delete, edit |
| Channels | 5 | create, join, leave, send, history |
| System Events | 2 | get_events, wait_for_event |
| Scheduler | 3 | add, remove, list |
| Auto-Reply | 3 | add, remove, list |
| Workflows | 4 | create, run, remove, list |
| Memory | 10 | store, search, recall, list, forget, stats, archive, clear, configure, rebuild |
| Utilities | 4 | bus_health, ask_hermes, hermes_send, hermes_channels |
All tools are prefixed with their category (e.g., agent_register, memory_search). Full schemas via MCP tools/list or bus help.
Utility Tools
Doctor
Diagnostics and health checking. Verifies bus directory integrity, agent registry validity, orphaned agents, stale PID files, message inbox sizes, channel consistency, schedule/rule/workflow validity, and event log integrity.
bus doctor [--quick] [--fix] [--report] [--watch]
18 diagnostic checks: directory existence, permissions, required subdirectories, agent profiles, orphaned agents, stale agents (>24h), message file integrity, channel metadata, schedule/rule/workflow validity, disk usage, stale PIDs, event log integrity.
Tunnel
Cross-machine bus proxy. Expose a local bus to a remote machine via HTTP REST with optional authentication and sync.
bus tunnel server --port 9090 --secret mytoken
bus tunnel client --host 192.168.1.100 --port 9090 --secret mytoken
bus tunnel sync --remote http://192.168.1.100:9090 --interval 5000
bus tunnel ssh --remote user@server.example.com
| Endpoint | Method | Description |
|---|---|---|
/health |
GET | Server status + agent count |
/bus/agents |
GET | List bus agents |
/bus/file/:path |
GET | Read a bus file |
/bus/send |
POST | Send message to bus |
/bus/write |
POST | Write a bus file |
/bus/register |
POST | Register an agent remotely |
Backup
Gzip-compressed backup of the entire .bus/ directory with SHA-256 checksum.
bus backup # Create timestamped backup
bus backup --list # List available backups
bus backup --info <file> # Show metadata & contents
bus backup --restore <file> # Restore (auto pre-restore backup)
bus backup --diff <file> # Compare with current state
bus backup --cleanup [days] # Remove old backups
bus backup --auto # Backup only if changes detected
bus backup --watch [minutes] # Auto-backup every N minutes
Performance
Environment: Node v24.12.0 · Windows Server · File-based (no database)
| Metric | Result |
|---|---|
| Message throughput | 1,082 msg/sec |
| Memory writes | 25,000/sec |
| Read inbox (25 msgs) | 0.54 ms/msg |
| List + filter 20 agents | <0.05 ms/agent |
| Keyword search (100 entries) | <0.01 ms/entry |
| Store 1 memory (JSONL append) | 40 μs |
File-based bus achieves these speeds because there is zero network overhead, no serialization beyond JSON.parse, and append-only JSONL for memory writes. The bottleneck on NTFS is directory listing at high file counts; ext4/apfs will perform even better.
Related
- MCP Protocol — Model Context Protocol specification
- Hermes Agent — MCP agent for OpenClaw
- Clew Code — Multi-provider AI coding agent
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.