| name | hermes-mcp-backend |
| description | Use when exposing Hermes Agent as an MCP server/backend so Claude Code, Codex, or another agent can query Hermes conversations, send messages, request approvals, fetch attachments, inspect channels, or delegate work through Hermes. |
Hermes MCP Backend
Use this when Hermes should act as a backend for other agents.
Core Use Cases
- Remote approval gate: another agent pauses risky work and Hermes sends an approval prompt to Discord/Telegram/Slack.
- Walk-away mode: long coding sessions report progress to your phone and accept short replies.
- Context bridge: Claude Code, Codex, or another tool can query Hermes conversations, channels, or memory without copying everything into context.
- Delegation backend: a local agent delegates research, messaging, or task tracking to Hermes.
Setup Flow
- Verify Hermes gateway works first.
- Expose Hermes through its MCP server mechanism or a small wrapper.
- Register the MCP server with the client agent.
- Test low-risk tools first:
- List conversations/channels.
- Fetch recent messages.
- Send a test message to a private test channel.
- Confirm auth, scope, and logs.
- Document the MCP command, host, ports, and restart procedure.
Client Verification
From the MCP client:
- List available Hermes tools.
- Fetch a known conversation.
- Read a few recent messages.
- Send a controlled test message.
- Ask the client agent to summarize what it learned from Hermes data.
Security Rules
- Do not expose Hermes MCP publicly without authentication.
- Prefer local SSH tunnel, private network, or tightly scoped credentials.
- Treat chat history as sensitive.
- Restrict send-message tools if the client agent is untrusted.
- Use approval gates for destructive or external-facing actions.
Good Tool Surfaces
list_conversationsget_messageslist_channelssend_messagefetch_attachmentrequest_approvalcreate_kanban_taskget_task_status
Source Notes
Distilled from the user-provided "seven levels of Hermes Agent" transcript, level 7.
