For AI assistants

This page is for agents, their AI assistants, and anyone who wants the exact rules. Avlira exposes a read-only MCP server. An agent authorises their own assistant through OAuth. The assistant can then read that agent's records and explain them. That is the whole contract.

What an agent's AI can do

With a connection the agent approved, the assistant can read and explain that agent's own records:

• Year summary: what they were paid this year, how many deals, whether the W-9 is on file
• Statements: their commission statements
• Deal breakdown: what a deal paid, the split applied, and every fee, to the cent
• Cap status: how much has been paid toward the cap, what remains, when it resets
• Dues status: what is owed, what is paid, and the pay link
• Their CDA and 1099 summary: as a short-lived signed download link, for their own deals only
• Onboarding status and office answers: which steps are done, and the brokerage's own answers to common questions
• Sign-in help: a one-time sign-in link when the agent is locked out, sent to the agent, never to the AI

Example questions: What did I net on the Maple Street closing? How much is left before my cap? Do I owe dues this month? I lost last year's 1099 email, can you get it?

What an agent's AI cannot do

• No writes. No posting, voiding, editing, approving or sending. The server exposes no tool that changes data.
• No money movement. It cannot take a payment, issue a payout, or change a plan.
• No other agents. One connection, one agent. No colleague statements and no office totals.
• No brokerage financials. Office revenue and other brokerages' data are not reachable.
• No W-9s or signed agreements. Only the agent's own CDA and 1099 summary, through a link that expires.
• No impersonation. An assistant cannot act as the office manager or activate an agent.

Read-only, and how that is enforced

There is one scope, read-only, and the tools behind it take no agent id: every call reads the signed-in agent's own records, so a prompt injection can at most read what that agent could already see. Every call is logged against the agent and the connection.

Start free with your AI assistant

You don't need to be technical. Add Avlira to your assistant once, then ask it in plain words.

In claude: Settings → Connectors → Add custom connector → paste https://app.avlira.ai/mcp → then say "Set up my brokerage on Avlira."

In chatgpt: Settings → Apps & Connectors → Advanced settings → turn on Developer mode → Create → paste https://app.avlira.ai/mcp → then say "Set up my brokerage on Avlira."

Your assistant asks for your brokerage name, your name and your email, creates your free back office, and gives you a link to open it. Then it can add your agents, your commission plan and your deals for you, and tell you what's waiting. Prefer a form? Start free here.

How an agent connects their own assistant

1. In the Avlira app, open Home → Your AI assistant.
2. Follow the Claude or ChatGPT steps shown there; the address to paste is on that screen with a Copy button.
3. Sign in with your texted code when your assistant asks, and tap Allow.

Home → Your AI assistant → Disconnect. Access ends immediately, and you get a notice whenever a new assistant connects.

Notes for assistant builders

• Transport is MCP over HTTPS; authorisation is OAuth with per-agent consent and a token that can be revoked at any time.
• Responses carry the amounts as recorded, with the transaction and date they belong to. There is no rounding applied for display beyond the stored value.
• If a question needs data outside the granted scopes, the correct behaviour is to tell the agent which scope to grant — not to guess.
• Rate limits and terms for programmatic access: hello@avlira.ai.