> ## Documentation Index
> Fetch the complete documentation index at: https://docs.auditynow.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Connect Claude

> Use Claude Desktop or claude.ai with Audity via npx @auditynow/connect (auto-configures) or manual MCP setup.

## Fastest way: one command

```bash theme={null}
npx @auditynow/connect
```

This auto-configures Claude Desktop and claude.ai for you. Skip to [Verify it works](#verify-it-works) below.

For details on flags, status, and manual setup for other tools, see [Connect with one command](/guides/connect-cli).

## Manual setup (if you prefer)

If you want to configure Claude manually or your tool wasn't auto-detected:

<Steps>
  <Step title="Get a Personal Access Token">
    Go to **Settings → API Tokens** in your Audity workspace and create a token with **Read + Write** scopes. It starts with `aky_` and is shown only once.
  </Step>

  <Step title="Open Connectors">
    Claude Desktop or claude.ai → Settings → **Connectors** → **Add Custom Connector**.
  </Step>

  <Step title="Enter the URL">
    ```
    https://app.auditynow.com/api/mcp
    ```
  </Step>

  <Step title="Configure auth">
    The connector needs `Authorization: Bearer aky_<your-token>` on every request.

    * If Claude shows an "Authorization header" or "API key" field: paste `Bearer aky_<your-token>` (the full string including `Bearer `, since Claude does not auto-prefix in this flow).
    * If Claude redirects you through an OAuth approval first: approve it. You'll still get a prompt for your Audity API key. Paste your `aky_<your-token>` (this time without `Bearer `, the connector wraps it).

    The exact UX depends on Claude Desktop's current connector flow. If you see something unexpected, check [Troubleshooting](#troubleshooting).
  </Step>

  <Step title="Test">
    Start a new chat. Ask:

    > List my Audity projects.

    Claude should call `audity_list_projects` and return your real projects. If it asks for permission first, approve.
  </Step>
</Steps>

## What you get

The MCP server exposes 100+ `audity_*` tools covering the full audit pipeline:

* **Navigation & project management**: `audity_get_navigation_status`, `audity_list_projects`, `audity_create_intake`, `audity_get_project_intake`
* **Document & interview intake**: `audity_upload_document`, `audity_create_interview_session`
* **Analysis**: `audity_enqueue_document_analysis`, `audity_enqueue_interview_analysis`, `audity_enqueue_audit_analysis`
* **Results**: `audity_get_project_opportunities`, `audity_create_roi_calculation`, `audity_select_opportunities`
* **Deliverables**: `audity_enqueue_pdf_generation`, `audity_enqueue_gamma_deck`, `audity_enqueue_stakeholder_memos`
* **Job management**: `audity_get_job`, `audity_list_jobs`, `audity_get_gamma_status`, `audity_get_pdf_status`
* **Leads & conversion**: `audity_list_leads`, `audity_convert_lead`
* **Nucleus (persistent memory)**: `audity_list_memories`, `audity_create_memory`
* **Account**: user, tier, and credits info

For the guided audit conductor (auto-orchestrates the pipeline), see [Guided Audit Conductor](/guides/audit-conductor).

## Verify it works

Ask Claude any of these:

```text theme={null}
List my Audity projects.
```

```text theme={null}
What's my Audity tier and how many credits do I have left?
```

```text theme={null}
Search my Audity Nucleus for anything related to {a real client of yours}.
```

If Claude returns real data (your projects, tier, memories), you're connected. If it says "I don't have access to Audity" or returns 401/403, jump to [Troubleshooting](#troubleshooting).

## Tips for agent prompts

* **Confirm before spending credits.** Credit-spending steps have approval gates; Claude will ask before running them. Project creation and lead conversion cost 1,000 credits each; audit analysis consumes AI compute credits on top of that. If you want Claude to auto-proceed, add "proceed without asking" to your prompt.
* **Cite opportunity and memory IDs.** Have Claude include the `id` field when summarizing. That way you can verify Claude isn't confabulating.
* **Use the `confidence` field.** Tell Claude to flag any memory under 0.8 confidence as a hypothesis, not a fact. AI-generated patterns are educated guesses.

## Troubleshooting

<AccordionGroup>
  <Accordion title="Claude says it can't find Audity tools">
    Check Settings → Connectors. If the Audity connector is red, re-authorize. If your PAT was revoked or expired, generate a new one in Settings → API Tokens and update the connector.
  </Accordion>

  <Accordion title="Every call returns 401 PAT_MALFORMED">
    Token format is wrong. Most common cause: the connector double-prefixed `Bearer Bearer aky_...`. Edit the connector auth and remove one `Bearer `, or paste only `aky_...` (depends on which UX field you're filling in Step 4 above).
  </Accordion>

  <Accordion title="Every call returns 403 PAT_SCOPE_INSUFFICIENT">
    Token lacks `write` scope. Generate a new token with both `read` and `write` scopes, then update the connector.
  </Accordion>

  <Accordion title="403 errors on every call but token looks fine">
    Your plan may not include the operation you're trying to run. Check `audity_get_tier` for your tier and read any `requiredTier` field in the 403 response.
  </Accordion>

  <Accordion title="429 rate limited mid-batch">
    See [Authentication → Rate limits](/authentication#rate-limits). Most common ceiling: write limits on batch operations. Consider serializing or adding delays.
  </Accordion>
</AccordionGroup>

## What's next

* [Run a full audit workflow →](/guides/running-an-audit)
* [Lead conversion playbook →](/guides/lead-conversion)
* [Working with Nucleus →](/guides/working-with-nucleus)