Set up in your harness

Smithers is driven by an agent, not clicked by a human. You ask your coding agent for an outcome and it reaches for Smithers to run the work, watch it, and clear approvals on your behalf. Setting up takes one of two paths: paste a prompt and let your agent wire everything, or run three commands yourself.

Fastest path: let your agent do it

Paste this into your harness of choice (Claude Code, Codex, Cursor, Copilot, or any agent with shell access). It installs the workflow pack, syncs Smithers’ generated CLI skill files where supported, registers the MCP server, then reports which starter fits your next task.

Setup prompt

Set up Smithers (a durable AI workflow orchestrator) in this repo for me.

1. Run `bunx smithers-orchestrator init` to scaffold `.smithers/`.
2. Sync Smithers skill files where supported and register its MCP server:
   `bunx smithers-orchestrator skills add`
   `bunx smithers-orchestrator mcp add`
3. Read `bunx smithers-orchestrator docs` so you know the CLI verbs and components.
4. Run `bunx smithers-orchestrator starters` and tell me which workflow fits the
   task I describe next.

From now on, when I ask for multi-step, long-running, or human-in-the-loop work,
use Smithers to orchestrate it instead of doing it inline.

init prints what it scaffolds, which coding agents it detected on your machine, and the next commands to try. skills add syncs the generated CLI skill set to supported skill locations; mcp add registers Smithers as an MCP server for detected MCP-capable agents such as Claude Code, Codex, Cursor, Gemini, OpenCode, Amp, Windsurf, and more.

Or wire it by hand

# 1. Scaffold the workflow pack into .smithers/
bunx smithers-orchestrator init

# 2. Sync the generated Smithers CLI skill set where supported
bunx smithers-orchestrator skills add

# 3. Register Smithers as an MCP server in every detected agent
bunx smithers-orchestrator mcp add

Scope the MCP install with --agent <name> for one harness, or use --no-global with skills to write into the current repo instead of your home directory:

bunx smithers-orchestrator mcp add --agent cursor
bunx smithers-orchestrator skills add --no-global

No skills directory in your agent? Point it at bunx smithers-orchestrator docs-full (prints the full docs bundle) or bunx smithers-orchestrator ask "<question>", and add a few standing lines to AGENTS.md or CLAUDE.md so it remembers Smithers is available.

Pick your harness

Each page covers the exact paths, config files, and any harness-specific steps.

Harness	Setup page
Claude Code	agents/claude-code
Codex	agents/codex
Cursor	agents/cursor
GitHub Copilot	agents/copilot
Pi	agents/pi
Hermes	agents/hermes
OpenClaw	agents/openclaw

The full support matrix and the three integration surfaces (skill, MCP server, standing instructions) live in Agent Support.

Verify it worked

Ask your agent to list the workflows; under the hood it runs:

bunx smithers-orchestrator workflow list
bunx smithers-orchestrator starters

If the MCP server registered, the agent can now call list_workflows, run_workflow, watch_run, and resolve_approval directly. See the MCP Server reference for every tool.

Next steps

Quickstart runs a seeded workflow immediately.
Installation covers the .smithers/ layout in detail.
Agent Support is the per-harness hub.

Agent Support

Codex

Start

Articles

Learn

Build Workflows

Run and Operate

Workflow Pack

Components

Integrations

Agent Support

Examples

Contributing

Set up in your harness

Fastest path: let your agent do it

Or wire it by hand

Pick your harness

Verify it worked

Next steps

​Fastest path: let your agent do it

​Or wire it by hand

​Pick your harness

​Verify it worked

​Next steps

Fastest path: let your agent do it

Or wire it by hand

Pick your harness

Verify it worked

Next steps