Quickstart

Get an AI agent talking to your Audity workspace in one command.

This is the agent path: run one command, auto-configure your tools, run real workflows from your AI assistant. If you’d rather call the API directly with curl or your own HTTP client, see API Quickstart.

Prerequisites

• An active Audity account with API access. A write-capable, audit-capable plan is recommended if you want the agent to create projects, convert leads, or spend credits.
• One of: Claude Desktop, Claude Code, Claude.ai (Pro+), ChatGPT (Plus for Custom GPTs), Cursor, n8n, or any MCP-compatible client

1. One command to connect

npx @auditynow/connect

This command:

1. Opens your browser to approve the connection
2. Mints an Audity Personal Access Token (aky_...)
3. Auto-configures Claude Desktop, Claude Code, and Cursor
4. Prints manual setup instructions for ChatGPT and n8n
5. Verifies the connection works

That’s it. You’re connected.

For scope flags, status checks, and more details, see Connect with one command →.

2. (Optional) Manual setup for ChatGPT or n8n

If you didn’t use connect, or you prefer manual setup:

Claude (auto-configured)

Already done by npx @auditynow/connect. Start chatting.

Full Claude guide →

ChatGPT

Create a Custom GPT, add an MCP connection to https://app.auditynow.com/api/mcp with your token. See connect’s manual output or ChatGPT guide →

Cursor (auto-configured)

Already done by npx @auditynow/connect. Start chatting.

Full Cursor guide →

n8n

Add an MCP Client Tool sub-node to your AI Agent. Endpoint https://app.auditynow.com/api/mcp. See connect’s manual output or n8n guide →

3. Verify it works

Ask your agent any of these:

List my Audity projects.

What's my Audity tier and how many credits do I have left this month?

Search my Audity Nucleus for anything related to {a real client of yours}.

If the agent returns real data, you’re connected. If it says “I don’t have access to Audity” or returns 401/403, check your platform’s guide for troubleshooting.

4. Run your first agent-driven audit

Try a task that chains multiple steps. Use a real client name or a hypothetical one:

Start a new Audity project for "Globex Industries", mid-market manufacturing.
Then run a comprehensive audit analysis. Once it's done, summarize the top three
opportunities ranked by impact and flag the highest-severity risk. Cite each
opportunity by ID.

Under the hood:

1. audity_create_project, creates the project (1,000 credits)
2. audity_upload_document / audity_create_interview_session, adds source material (MCP-only tools; a comprehensive audit needs document and interview analyses first)
3. audity_enqueue_document_analysis / audity_enqueue_interview_analysis, then audity_enqueue_audit_analysis, enqueues the synthesis (~10-15 minutes)
4. The agent polls audity_get_job until status: "completed"
5. audity_get_project_opportunities, fetches the AI-generated opportunities, and the agent synthesizes a summary citing the IDs

The whole loop runs in the agent’s context. Audity logs every call, deducts credits, applies audit logging, RLS, and all security checks. The Guided Audit Conductor handles this ordering and the polling for you.

Save the project ID. The agent will mention it. You’ll need it for follow-ups like “pull the deliverables for “.

5. Try a Nucleus workflow

Audity’s persistent memory layer is the agent’s other superpower. After running a few audits, try:

What patterns do I have stored in Audity Nucleus across my healthcare clients?
For each one, include the confidence score and source type.

The agent calls GET /api/nucleus/memories?type=pattern and surfaces what’s there. If you don’t have any patterns yet, ask:

Capture this note for me into Audity Nucleus: "Healthcare clients consistently
underestimate the data governance lift required before any AI deployment.
Plan a 4-6 week governance phase before model work."

The agent calls POST /api/nucleus/memories and creates an explicit memory you’ll find on every future Nucleus query.

What MCP unlocks (vs REST-only)

Connected via MCP, your agent can do things a REST-only client cannot: upload documents (audity_upload_document), create interview sessions, poll PDF-generation status, trigger deliverable generation (PDF, Gamma deck, stakeholder memos), and enqueue web research. See REST vs MCP.

Two things stay browser-only: streaming chat-style responses (Audity returns structured JSON) and PAT management (tokens are created and revoked from a browser session, not via the API).

If a workflow you need crosses one of these, tell us.

What’s next

Run a full audit workflow

The end-to-end agent recipe for taking a client from intake to deliverables.

Lead conversion playbook

Triage your ReadyLink inbox and convert qualified leads from your AI assistant.

Working with Nucleus

Memories, captures, insights, contacts. The persistent layer.

Authentication deep dive

Token formats, scopes, rotation, error codes, what’s logged.

API Quickstart (curl)

Skip MCP, call the API directly from a terminal or script.

Browse the spec

The full OpenAPI 3.1 reference. Click any endpoint to try it live.