Set Up AIntent in 30 Seconds

Copy the message below and paste it into your AI tool (Claude Code, Codex).

Complete ALL steps below. Do NOT skip, summarize, or combine steps.

Step 1 — Check API Key

Ask the user if they have an AIntent API key.

If they do not have one, instruct them to:

1. Go to https://aintent.dev and sign up
2. After signing in, go to the Auth / API Keys page in the dashboard
3. Click "Create New Key" and copy the generated key
4. Come back here when ready

Step 2 — Enter API Key

Present BOTH options to the user:

export AINTENT_API_KEY=your-key-here

Step 3 — Configure MCP Server

Detect which CLI tool the user is running and configure accordingly.

{
  "mcpServers": {
    "aintent": {
      "type": "http",
      "url": "https://aintent.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

[mcp_servers.aintent]
url = "https://aintent.dev/api/mcp"
http_headers = { Authorization = "Bearer YOUR_API_KEY" }

Wait for the user to confirm completion. Do NOT proceed until confirmed.

Step 4 — Install Skill File

The /aintent command requires a skill file. Download for the user's CLI tool:

mkdir -p ~/.claude/commands
curl -o ~/.claude/commands/aintent.md https://aintent.dev/skill/SKILL.md

mkdir -p .agents/skills/aintent
curl -o .agents/skills/aintent/SKILL.md https://aintent.dev/skill/SKILL.md

This adds the /aintent command.

Step 5 — Verify Connection

After installation:

1. Start a new session in the configured tool
2. Check that AIntent appears as a connected MCP server
3. Report the connection status to the user

Step 6 — Start Using AIntent

You MUST display the following to the user in their preferred language. Do NOT summarize or skip any mode.

"AIntent is now installed. Here's how it works:

• AIntent is OFF by default. It does not interfere with your normal workflow.
• When you activate it, AIntent will ask clarifying questions to understand your intent before generating output.
• Once the task is complete, AIntent automatically turns OFF again."

Ask the user which mode they would like to start with:

• off — AIntent is inactive. Your AI tool operates as provided by its service provider.
• quick — Light check. 1–2 clarifying questions before proceeding.
• standard (Recommended) — Collects intent thoroughly before generating output.
• deep — In-depth. For architecture decisions, complex planning, or high-stakes tasks.

Once the user chooses, run:

/aintent [mode]

Example: /aintent standard

After the task is complete, AIntent automatically returns to off mode.

Mid-Session Disengage

Users can disengage (turn off) AIntent mid-session by running /aintent off. When this happens, all accumulated context — interview questions and answers, slot coverage, and confidence data — is dumped to the LLM at once. This allows the LLM to leverage the collected insights for the remainder of the session without further gating.