MCP Agent Bridge
A bridge server that enables Hermes Agent and OpenClaw to share tools and memory, forming a cooperative multi-agent system.
README
MCP Agent Bridge
A lightweight MCP (Model Context Protocol) bridge server that enables Hermes Agent and OpenClaw to share tools and memory, forming a cooperative multi-agent system.
Architecture
┌─────────────┐ ┌──────────────────┐ ┌─────────────┐
│ Hermes │ MCP │ Agent Bridge │ CLI/ │ OpenClaw │
│ Agent │◄───────►│ (FastMCP SSE) │◄───────►│ CLI │
│ │ client │ Port :18900 │ HTTP │ │
│ Feishu/WX │ │ │ │ Image/Web │
│ Memory/Plan │ │ Shared Memory │ │ TTS/Video │
└─────────────┘ └──────────────────┘ └─────────────┘
Hermes contributes messaging channels (Feishu, WeChat, Telegram…) and its planning/reasoning capabilities.
OpenClaw contributes AI image generation, web search, TTS, and video generation.
The Bridge makes each agent's strengths available to the other through standard MCP tool calls, plus a shared SQLite-backed memory store.
Exposed Tools
| Tool | Direction | Description |
|---|---|---|
oc_image_generate |
OC → Hermes | AI image generation via OpenClaw |
oc_web_search |
OC → Hermes | Web search via OpenClaw |
oc_web_fetch |
OC → Hermes | URL content extraction via OpenClaw |
oc_tts_convert |
OC → Hermes | Text-to-speech via OpenClaw |
hermes_send_message |
Hermes → OC | Send messages through Hermes's platforms |
hermes_chat |
Hermes → OC | Delegate tasks to Hermes's AI agent |
shared_memory_read |
Bidirectional | Read from shared key-value store |
shared_memory_write |
Bidirectional | Write to shared key-value store |
shared_memory_list_keys |
Bidirectional | List memory keys with optional filters |
shared_memory_delete |
Bidirectional | Delete a memory key |
bridge_health |
— | Health check and registered module list |
Quick Start
Prerequisites
- Python 3.11+
- Hermes Agent with API Server enabled
- OpenClaw CLI installed and configured
pip install "mcp[cli]>=1.0" aiohttp pyyaml
Install
git clone https://github.com/fkdt01/mcp-agent-bridge.git
cd mcp-agent-bridge
pip install -e .
Configure
cp config.example.yaml config.yaml
# Edit config.yaml — fill in your API keys and paths
Key settings:
openclaw:
cli_path: "openclaw" # or full path like /usr/local/bin/openclaw
image_provider: "openai" # default image gen provider
web_search_provider: "duckduckgo"
hermes:
api_url: "http://127.0.0.1:8888"
api_key: "" # or set HERMES_API_KEY env var
default_channel: "feishu"
memory:
backend: "sqlite"
db_path: "data/bridge_memory.db"
Run
# Start the bridge server
python -m bridge.server
# With custom options
python -m bridge.server --config /path/to/config.yaml --port 18900 --log-level DEBUG
Connect Hermes
Add to ~/.hermes/config.yaml:
mcp_servers:
agent-bridge:
transport: sse
url: http://127.0.0.1:18900/sse
Connect OpenClaw
Add to ~/.openclaw/openclaw.json → mcpServers:
{
"mcpServers": {
"agent-bridge": {
"transport": "sse",
"url": "http://127.0.0.1:18900/sse"
}
}
}
Usage Examples
Hermes generates an image via OpenClaw
When Hermes needs to generate an image (e.g. from a Feishu chat), it calls:
oc_image_generate({
prompt: "A futuristic city at sunset, cyberpunk style",
aspect_ratio: "16:9",
model: "openai"
})
→ Bridge runs openclaw capability image generate --prompt "..." --json
→ Returns image path/metadata to Hermes
→ Hermes delivers the image to the user
OpenClaw sends a Feishu message via Hermes
hermes_send_message({
message: "✅ Image generation complete!",
target: "feishu"
})
→ Bridge POSTs to Hermes API /v1/chat/completions
→ Hermes delivers the message to the Feishu channel
Shared memory between agents
Hermes writes project context:
shared_memory_write({
key: "project.math-game.pet-design",
value: "算术小喵: 草原主题, 绿色配色",
source: "hermes"
})
OpenClaw reads it later:
shared_memory_read({ key: "project.math-game.pet-design" })
→ { value: "算术小喵: 草原主题, 绿色配色", source: "hermes", updated_at: 1745631000 }
Adding Custom Tools
Create a new file in bridge/tools/ following the naming convention:
oc_*.py— OpenClaw-backed toolshermes_*.py— Hermes-backed toolsshared_*.py— Shared/bidirectional tools
Each module must expose a register(mcp, config) function:
"""My custom tool."""
from mcp.server.fastmcp import FastMCP
from typing import Any
def register(mcp: FastMCP, config: dict[str, Any]) -> None:
@mcp.tool()
async def my_custom_tool(param: str) -> dict[str, Any]:
"""Tool description — this becomes the MCP tool description."""
# Your implementation here
return {"result": "ok"}
The tool is auto-discovered and registered on server startup.
Running as a systemd Service
cat > ~/.config/systemd/user/mcp-agent-bridge.service << 'EOF'
[Unit]
Description=MCP Agent Bridge Server
After=network.target
[Service]
Type=simple
WorkingDirectory=/path/to/mcp-agent-bridge
ExecStart=/usr/bin/python -m bridge.server
Restart=on-failure
RestartSec=5
Environment=HERMES_API_KEY=your_key_here
[Install]
WantedBy=default.target
EOF
systemctl --user daemon-reload
systemctl --user enable --now mcp-agent-bridge
Project Structure
mcp-agent-bridge/
├── bridge/
│ ├── __init__.py
│ ├── __main__.py
│ ├── server.py # FastMCP server entry point
│ ├── config.py # YAML + env-var config loader
│ ├── openclaw_runner.py # Shared async CLI subprocess runner
│ ├── memory.py # SQLite-backed shared memory store
│ └── tools/
│ ├── __init__.py # Auto-discovery & registration
│ ├── oc_image.py # Image generation
│ ├── oc_web.py # Web search & fetch
│ ├── oc_tts.py # Text-to-speech
│ ├── hermes_messaging.py # Message sending
│ ├── hermes_chat.py # Task delegation
│ └── shared_memory.py # Shared memory tools
├── config.example.yaml
├── .gitignore
├── LICENSE
├── README.md
└── pyproject.toml
License
MIT
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
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
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.