Hermes Agent Integration Guide
Plug-and-play persistent memory for Hermes Agent. Add one MCP server and your agent remembers everything across sessions, platforms, and conversations.
Overview
Hermes Agent is an AI agent framework that supports the Model Context Protocol (MCP) for extensible tool use. AgentRecall provides a memory MCP server that automatically stores and retrieves contextual information — preferences, corrections, facts, and temporal events — so your Hermes agent carries knowledge forward between sessions.
The integration is a single YAML config change and a restart. No code modifications needed.
Prerequisites
- Python 3.10 or later installed on your system.
- Hermes Agent installed and operational (hermes --version should work).
- An AgentRecall account with an active API key (ark_...).
Installation
Install the AgentRecall MCP server package globally:
pip install agentrecall-mcpSetup
Open your Hermes configuration file and add the MCP server block:
"comment"># Edit the config file\nvim ~/.hermes/config.yamlAdd the following section to your config:
"keyword">mcp_servers:\n agentrecall:\n command: "agentrecall-mcp"\n env:\n AGENTRECALL_API_KEY: "ark_YOUR_KEY"\n AGENTRECALL_AGENT_ID: "hermes-agent"Replace ark_YOUR_KEY with the API key from your AgentRecall dashboard. The AGENTRECALL_AGENT_ID value determines the namespace for all stored memories.
Restart Hermes
Apply the configuration by restarting the Hermes gateway:
hermes gateway restartThis shuts down the current session, reloads config.yaml, and spawns the MCP servers defined in the configuration.
Verify Installation
Check that the MCP server is registered:
hermes mcp listYou should see `agentrecall` listed with a healthy status.
You can also verify it works end-to-end by asking your agent:
"comment"># In a Hermes conversation:\n"Search my memories for my preferred editor"If the agent returns results (or says it found nothing yet), the integration is working.
Configuration Reference
All configuration is passed as environment variables to the MCP server process. Set them in the env block of your config.
| Variable | Description | Required | Default |
|---|---|---|---|
| AGENTRECALL_API_KEY | Your AgentRecall API key (ark_...) | Yes | — |
| AGENTRECALL_AGENT_ID | Agent identifier (e.g. hermes-agent) | Yes | hermes |
| AGENTRECALL_BASE_URL | Custom API endpoint | No | https://api.agentrecall.com |
| AGENTRECALL_LOG_LEVEL | Logging verbosity | No | info |
Use Cases
Session Memory
Continue conversations exactly where you left off. The agent recalls what you discussed, decisions you made, and tasks that are still pending.
Preference Tracking
Over time the agent learns your coding style, communication preferences, tool choices, and workflows — applying them without being asked.
Multi-Platform Context
Interact with Hermes from your CLI, Telegram, Discord, or Slack and the memory is shared. Context flows seamlessly between platforms.
Troubleshooting
agentrecall-mcp with your system Python, Hermes won't find the binary.Solution A: Install inside Hermes's venv:
"comment"># Find Hermes venv location\nhermes venv path\n\n# Install into it\n$(hermes venv path)/bin/pip install agentrecall-mcp"keyword">mcp_servers:\n agentrecall:\n command: "/path/to/agentrecall-mcp"\n env:\n AGENTRECALL_API_KEY: "ark_YOUR_KEY"\n AGENTRECALL_AGENT_ID: "hermes-agent"Cleanup:
"comment"># Find orphaned agentrecall processes\nps aux | grep agentrecall-mcp\n\n# Kill them\nps aux | grep agentrecall-mcp | grep -v grep | awk '{print $2}' | xargs kill"comment"># Update globally\npip install --upgrade agentrecall-mcp\n\n# Or update inside Hermes venv\n$(hermes venv path)/bin/pip install --upgrade agentrecall-mcp\n\n# Restart to pick up the new version\nhermes gateway restart