The Centered CLI
centered turns our agency into something you can call like an API. Request an audit, buy a sprint, approve a deliverable — in the terminal you're already in, or from the AI assistant already sitting next to you, over MCP.
No npm required — just Node 18+. Prefer a package manager?npm install -g @centered/cli, or run it once withnpx @centered/cli. Update anytime withcentered upgrade.
$ centered login
→ opening your browser… signed in as you@yourcompany.com
$ centered init
→ detected repo · project connected
$ centered add sprint --goal "Finish team accounts and permissions"
→ two-week outcome sprint · senior-led
✓ CS-1842 submitted — Centered will confirm scope and price
$ centered status
Engagement Active · Sprint 3 of 4
In progress Billing migration · Admin dashboard
Awaiting you CS-1842 (proposal)Audits, features, fixes, estimates, rescues. Each command collects the context we actually need, submits it, and hands you a request number. The team is notified immediately.
$ centered request audit$ centered request feature$ centered request fix$ centered request estimate$ centered request rescueProvision engineering the way you provision infrastructure. Sprints and hour blocks are fixed-price and can be bought on the spot — checkout opens in your browser and the CLI confirms once payment clears. Developers and teams come back as a proposal.
$ centered add sprint$ centered add hours 40$ centered add developer$ centered add team$ centered add sprint --goal "Ship team accounts" --purchaseSee your engagement, what's in flight, and what's waiting on you. Approve a proposal or deliverable to move it forward, or decline with a reason to send it back for changes.
$ centered status$ centered review$ centered approve CS-1842$ centered decline CS-1842Message the team, check what you've bought against what you've used, pull invoices, or put a human on the calendar — without opening a portal, a spreadsheet, or an email thread.
$ centered message "Ship onboarding first"$ centered billing$ centered callMCP server
Connect Centered to your assistant and it can request work, price capacity, check status, and hand you a checkout link. Works with Claude, Claude Code, Cursor, OpenClaw, Hermes — anything that speaks MCP or runs a terminal.
Two questions, then the exact steps for your assistant. No technical background needed.
Copy one prompt
Paste this into Claude Code, Cursor, or any assistant that can run commands. It installs the CLI, connects the skill, and signs you in — then it can hire us for you.
Install and connect Centered Software: runcurl -fsSL https://api.centeredsoftware.com/install.sh | sh, thencentered skill install, then centered login. Then help me request work and check our engagement status.
Prefer no terminal? (Claude)
One click adds Centered to Claude's browser or desktop app — just confirm and sign in. (Custom connectors need a paid Claude plan.)
Add Centered to ClaudeOr add it manually — Settings → Connectors → "Add custom connector," and paste:
# sign in once — the server reuses your CLI session
$ centered login
# Claude Code
$ claude mcp add centered -- npx -y @centered/mcpClaude Desktop or Cursor — add the same command undermcpServers:
{
"mcpServers": {
"centered": {
"command": "npx",
"args": ["-y", "@centered/mcp"]
}
}
}Then just say the thing
Ask Centered for a security audit before our SOC 2 review.
Buy a sprint to finish team accounts and permissions.
What's the status of our engagement?
Approve CS-1842.
Fifteen tools
request_service · list_offeringsadd_sprint · add_hours · add_developer · add_teamget_status · get_billing · whoamilist_reviews · approve · request_changessend_message · schedule_callcreate_checkoutGuardrails, built in
Nothing here spends your money on its own. Buying tools return a secure checkout link that you open and complete yourself — an assistant can price a sprint, but it cannot charge you for one. Every submission reports its request number back to you, and anything ambiguous or high-stakes routes to a call with a human.
Prefer a skill? It's in the box.
If you're in Claude Code and would rather your agent drive the CLI directly, runcentered skill install — it writes the Centered skill to ~/.claude/skills/, or add--project to scope it to a single codebase. Same capabilities, same guardrails; MCP is the better fit outside the terminal.
Every read command supports --json. Requests and scheduling run entirely on flags — no prompts to answer, and--no-browser prints URLs instead of opening one. Stable exit codes, so CI can act on the result.
The CLI and the MCP server submit structured requests. Neither reads, uploads, or analyzes your repository. Credentials are stored locally and are never printed.
Behind every command is the same senior team that builds everything else we ship. The terminal is the interface, not the engineer.
Sixty seconds
$ curl -fsSL https://api.centeredsoftware.com/install.sh | sh
$ centered login
$ centered request audit