Skip to main content
Hermes (Nous Research) is a self-improving, multi-channel agent with a terminal TUI, an OpenAI-compatible HTTP API, and first-class MCP support in both directions. Because Hermes is a native MCP client, the cleanest integration is to register Smithers as an MCP server. Hermes can then list, run, watch, and approve workflows as tools.

​
Register the MCP server

The fastest path is mcp add, which writes its server straight into Hermes’s ~/.hermes/config.yaml: 
bunx smithers-orchestrator mcp add --agent hermes
To do it by hand, Hermes reads MCP servers from ~/.hermes/config.yaml under mcp_servers. Add Smithers as a stdio server: 
mcp_servers:
  smithers:
    command: bunx
    args: ["smithers-orchestrator", "--mcp"]
Or use the Hermes CLI: 
hermes mcp   # browse the catalog / install / configure
Smithers’ tools surface as mcp_smithers_<tool>: mcp_smithers_run_workflow, mcp_smithers_watch_run, mcp_smithers_resolve_approval, and the rest. Full reference: MCP Server.

​
Install the skill

Hermes supports the SKILL.md (agentskills.io) format. Drop the onboarding skill into its skills directory. skills add also writes the generated set to the canonical ~/.agents/skills/, which Hermes can be pointed at: 
mkdir -p ~/.agents/skills/smithers
curl -fsSL https://raw.githubusercontent.com/smithersai/smithers/main/skills/smithers/SKILL.md \
  -o ~/.agents/skills/smithers/SKILL.md
curl -fsSL https://smithers.sh/llms-full.txt \
  -o ~/.agents/skills/smithers/llms-full.txt
Because Hermes is a native MCP client, the MCP server above is the primary integration; the skill is a complement, not a requirement.

​
Standing instructions

Hermes resolves an instruction file by priority (.hermes.md, then AGENTS.md, then CLAUDE.md), and the first match wins. Add an orchestration section to whichever the repo already uses: 
## Orchestration
Use Smithers for durable, multi-step, or human-in-the-loop agent work.
`bunx smithers-orchestrator starters` lists ready-made workflows.

​
See also