Hermes Agent + Claude Code

Why Pair Them?

• One Anthropic API key, two surfaces — Hermes on chat, Claude Code in the terminal.
• Same MCP servers (Linear, GitHub, Slack) accessible from both.
• Claude Code can run shell commands that drive Hermes deploys (PR comments, CI triggers).
• Hermes can ping you in chat when long-running Claude Code tasks finish.

Step 1: One Anthropic Key, Two Tools

Generate an Anthropic API key at console.anthropic.com. Set it in both places:

# For Claude Code (in shell)
export ANTHROPIC_API_KEY=sk-ant-...

# For Hermes Agent (in ~/.hermes/config.json)
{
  "models": {
    "providers": {
      "anthropic": {
        "apiKey": "sk-ant-..."
      }
    }
  },
  "agents": {
    "defaults": {
      "model": { "primary": "anthropic/claude-sonnet-4-6" }
    }
  }
}

Now both tools call the same model, share quota, and show up on the same Anthropic usage dashboard.

Step 2: Share MCP Servers

Claude Code reads MCP servers from ~/.claude/mcp.json and project-level .mcp.json. Hermes Agent reads them from ~/.hermes/mcp.json. The format is the same — symlink one to the other so a single edit reaches both:

# Symlink Hermes config to Claude's
mkdir -p ~/.hermes
ln -sf ~/.claude/mcp.json ~/.hermes/mcp.json

# Both tools now see the same MCP servers
ls -la ~/.claude/mcp.json ~/.hermes/mcp.json

Add a server once, both Claude Code and Hermes can use it. Linear, GitHub, Slack, Notion — same set on both sides.

Step 3: Let Claude Code Drive Hermes Deploys

If you self-host Hermes on a VPS, give Claude Code an SSH shortcut and a slash command that pulls + restarts the Hermes container:

# ~/.claude/commands/redeploy-hermes.md
Pull latest Hermes container on the VPS and restart it.

```bash
ssh hermes-vps "cd ~/hermes && git pull && docker compose up -d --build"
```

Now /redeploy-hermes in Claude Code restarts your Hermes bot in one keystroke.

Step 4: Have Hermes Ping You When Claude Code Finishes

Long-running Claude Code tasks (big refactors, full test suites) can write a webhook to your Hermes gateway and Hermes will DM you on Telegram when they finish:

# Claude Code hook in ~/.claude/settings.json
{
  "hooks": {
    "Stop": [
      "curl -X POST https://hermes.example.com/notify \
        -H 'Authorization: Bearer YOUR_HERMES_TOKEN' \
        -d '{\"text\": \"Claude Code finished in $(pwd)\"}'"
    ]
  }
}

Hermes routes the webhook to your Telegram chat. Walk away from the terminal — you'll get a chat ping when the work is done.

Step 5: Have Claude Code Read Hermes Memory

Hermes's long-term memory store is a JSON file. Expose it as a read-only MCP server so Claude Code can quote it during pair-programming:

# ~/.claude/mcp.json
{
  "mcpServers": {
    "hermes-memory": {
      "command": "cat",
      "args": ["~/.hermes/memory/long-term.json"]
    }
  }
}

Now Claude Code can answer “what did the customer tell Hermes about their deploy preference last week?” without you re-typing it.

Common Pitfalls

• Quota: sharing one Anthropic key means both tools consume the same usage. Set a budget alert in the Anthropic console.
• Tool name collisions: if both Claude Code and Hermes register a tool called send_email, the MCP server's namespace prefix avoids ambiguity. Use unique names per server.
• Memory leakage: Hermes's memory store can contain user PII. Don't expose it as an MCP server on a shared machine.

When to Use Each

See the side-by-side comparison at Hermes Agent vs Claude Code for which workflow each tool wins. Short version: Hermes for multi-user chat with memory, Claude Code for repo-aware code edits, both for everything in between.

Skip the Setup — Managed Hermes