@chieflab/cli

Hosted MCP from your terminal. ChiefLab's repo-aware launch flow plus approval-gated execution.

Package status: live on npm as @chieflab/cli.

curl -X POST https://api.chieflab.io/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Why a CLI when the MCP is an HTTP endpoint?

Because curl is great until you need the second call. Once you've got a runId to look up, an actionId to publish, a reviewUrl to open, an accountId to pass — you want a real shell tool, not a fistful of JSON-RPC envelopes.

This CLI:

• Saves your API key once (~/.chieflab/config.json, chmod 600)
• Wraps every operator + the closed-loop launch flow with sane flags
• Has zero npm dependencies (node 20+ built-ins only — install in <5s)
• Is honest about JSON output (--json on any command)

Install

Use the one-shot attach path from the repo you just built:

npx -y @chieflab/cli attach --runtime auto --project . --intent "launch this"

attach detects the local agent runtime, mints the workspace key, writes MCP config, plants repo instructions, and persists the pending intent so the agent can resume after a runtime restart. Both chieflab and chiefmo work as bin names for manual use later.

Auth

chieflab login                 # interactive prompt
chieflab login --key clp_...   # non-interactive
chieflab whoami                # confirm workspace
chieflab logout                # forget the key

Or skip the file and pass via env (good for CI):

export CHIEFLAB_API_KEY=clp_live_...
chieflab whoami

Get a key at https://chieflab.io/get-key.

The closed loop

# 0. Install repo instructions so agents call ChiefLab after build
chieflab init

# 1. Plan + draft from the current repo
chieflab launch-here https://yoursite.com --channels linkedin,x,email
#   → returns launchId + publishActions + reviewUrl

# 2. Approve on the reviewUrl (open it in a browser)

# 3. Fire the approved actions
chieflab publish launch-action-run_xxx-linkedin \
  --content "..." --platform linkedin --account zer_acc_xxx

chieflab send-email launch-action-run_xxx-email \
  --from "Brand <[email protected]>" --to [email protected] \
  --subject "Launch day" --html "<h1>...</h1>"

# 4. Wait 24h, then read results + get the next-move recommendation
chieflab review run_xxx

Full launch wedge in 4 commands.

init writes AGENTS.md, a Cursor rule, Claude command, Codex note, and an MCP config stub. It tells agents to call chieflab_get_users_after_build, surface agentGuide.userMessage, wait for the reviewUrl, execute only approved actions, resume with chiefmo_continue_launch_loop, and measure after 24h.

Compatibility: chieflab install-agent-hook still works. Public docs should prefer npx @chieflab/cli init.

chieflab init --dry-run
chieflab init --runtime cursor
chieflab init --runtime all --force

Launch Commands

chieflab launch-here https://yoursite.com  # preferred: gathers repo context
chieflab launch https://yoursite.com       # URL-only fallback
chieflab marketing "Diagnose our current funnel"

Each accepts --url <product-url> (auto-extract brand context), --tenant-id <id> (use stored tenant), --idempotency-key <key> (cached re-runs).

Trigger phrases

When an agent calls the CLI on a user's behalf, these phrases map to launch-here:

• "launch this"
• "get users"
• "market this"
• "announce this"
• "post this"

For an existing program ("why is X not working"), use chieflab marketing "<goal>" (calls chiefmo_diagnose_marketing).

Connections

chieflab connections           # show what's connected
chieflab connect ga4           # OAuth start (also: search_console, hubspot)
chieflab set-key zernio sk_... # for set-key flows (also: resend)

Inbox

chieflab runs                  # list workspace runs
chieflab runs --limit 50
chieflab open run_xxx          # print the signed reviewUrl

Tools introspection

chieflab tools                 # list the live MCP tools available on this key
chieflab tools --json          # for piping into jq

Flags

| Flag | What | |------|------| | --json | Emit raw JSON (suppresses pretty terminal output) | | --key <key> | Override saved key for this call only | | --endpoint <url> | Override endpoint (default https://api.chieflab.io/api/mcp) |

| Env | What | |-----|------| | CHIEFLAB_API_KEY | Wins over saved config (CI-friendly) | | CHIEFLAB_ENDPOINT | Wins over saved endpoint | | NO_COLOR | Disable ANSI colors |

License

Apache 2.0. Source at https://github.com/bdentech/chieflab.