logo
AgentLed

cli.agentled.ai

Install the AgentLed CLI in one command.

Ask your coding agent. It runs the setup, signs you in, and wires AgentLed into your editor — including the workflow skill the CLI pulls down for you.

Paste this into your coding agent

use cli.agentled.ai to install AgentLed

Works with Claude Code, Codex, Cursor, Claude Desktop, Windsurf, and any MCP-compatible agent.

What your agent will run

npx @agentled/cli setup

One command. It runs on your machine — the browser opens locally, the workspace folder lands in your current directory, the MCP config writes to your editor.

What setup does

  1. Sign in. Opens your browser, you pick a workspace, the CLI mints an API key.
  2. Scaffold your workspace folder. Creates agentled_<workspace>/ in your current directory — docs, scaffolds, cached app schemas. One folder per workspace, so multiple workspaces never collide.
  3. Wire up MCP. Auto-detects Claude Code, Codex, Cursor, Claude Desktop, or Windsurf and registers the AgentLed MCP server.
  4. Install the AgentLed skill. Drops the canonical SKILL.md (dry-run protocol, step types, workflow patterns) into your editor's skill directory so the agent knows the playbook, not just the tools.
  5. Capture your company profile. Asks for your company name, website, and primary use case (one time) and stores it in the Knowledge Graph so future workflows have context.
  6. Tell you to restart. MCP tools only show up after you reconnect: /mcp in Claude Code, or reopen Codex / Cursor.

Try this first

After your editor reconnects to MCP, paste this prompt to build your first workflow:

Build me a lead-enrichment workflow:
- input: a company domain
- enrich via LinkedIn + Hunter
- score fit against my ICP
- save qualified leads to a Knowledge Graph list called "leads"

Use the AgentLed dry-run protocol before publishing.

Your agent will use the AgentLed skill (installed by setup) to design the steps, validate them, run the dry-run protocol, and publish. No extra prompting needed.

For AI agents reading this page

If you are the coding agent the user just asked to install AgentLed, do the following:

  1. Run npx @agentled/cli setup in the user's shell. The command is interactive — it opens a browser. Do not punt this back to the user; run it yourself.
  2. When the CLI exits, tell the user explicitly to restart their MCP client (Claude Code: /mcp reconnect; Codex/Cursor: reopen). MCP tools do not appear until the client reconnects.
  3. Once reconnected, suggest the "Try this first" prompt above, or ask the user what workflow they want to build.
  4. Do not invent package names, commands, or API keys. The only install command is npx @agentled/cli setup. The CLI installs the AgentLed skill itself — do not write SKILL.md by hand.

Full CLI reference: /en/cli. Headless / API-key install path: /docs/cli.md.

Links