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

# MCP

> Connect Aura to Claude and ChatGPT so your assistant can answer questions about your calls, pipeline and revenue.

# Aura MCP Connector

Connect Aura to Claude or ChatGPT and ask about your own data in plain language:

* *"How did we do last week?"*
* *"Why are we losing deals?"*
* *"Prep me for my 2pm."*

The connector is **read-only**. It cannot book, cancel or reschedule calls, cannot
change a lead, and cannot record a payment. Nothing it does sends an email or a
calendar invite to anyone.

## Connect

**Endpoint**

```
https://api.aura-app.ai/mcp
```

### Claude

1. **Settings → Connectors → Add custom connector**
2. Paste the URL above
3. Sign in with your Aura account when prompted
4. If you belong to more than one Aura workspace, pick the one you want on the
   consent screen

Works in Claude on the web, desktop, and mobile.

### ChatGPT

1. **Settings → Connectors → Add**  (developer mode may be required depending on
   your plan)
2. Paste the URL above
3. Sign in with your Aura account

## What it can see

Exactly what you can see, and nothing else. Access is decided by your
membership of an Aura workspace and your role in it — the same rules as the
dashboard. Membership is re-checked on **every** request, so removing someone
from a workspace takes effect immediately rather than whenever their session
expires.

| Tool | What you'd ask |
| - | - |
| `prepare_for_call` | "Brief me on my next call" |
| `get_dashboard_analytics` | "How did we do last week?" |
| `diagnose_losses` | "Why are we losing deals?" |
| `get_attribution` | "Which campaigns actually produce revenue?" |
| `get_team_performance` | "How is each rep doing?" |
| `get_pipeline` | "What's in the pipeline?" |
| `find_calls`, `get_call_detail` | "Show me that call with Acme" |
| `get_lead_detail`, `search` | "Look up [ada@example.com](mailto:ada@example.com)" |
| `list_organizations` | Which workspaces you belong to |

Call recordings and transcripts are **never** returned.

## Things worth knowing

**Search is narrow on purpose.** It matches an email address exactly, or the
start of a name. A descriptive query like *"enterprise deals in Texas"* will not
match anything — use `find_calls` or the analytics tools for that.

**Totals come from the analytics tools.** If you ask for a list of calls you get
a page of them, not all of them, and the assistant is told to say so. Ask for
period totals rather than adding up a list.

**Data can be a few minutes stale.** Aggregate answers are cached for 5–10
minutes so repeated questions stay fast. Anything about a specific call or lead
is always live.

**There is a fair-use budget.** An assistant that queries in a tight loop will
be asked to slow down and reuse what it already has. Normal use will not come
close.

## Multiple workspaces

If you belong to more than one Aura workspace, pick one on the consent screen
when you connect. You can also say *"list my Aura workspaces"* and then name the
one you want.

If a question is ambiguous, the assistant is told which workspaces you belong to
and asked to pick — it will never silently guess.

## Troubleshooting

**"Couldn't reach the MCP server"** — usually a discovery problem rather than an
outage. Check `https://api.aura-app.ai/health` returns `200`. If it does,
disconnect and reconnect the connector.

**"Not enabled for this account"** — the connector is rolling out to workspaces
in stages. Contact Aura to be added.

**"You belong to more than one Aura workspace"** — expected. Name the workspace
you want, or reconnect and pick it on the consent screen.

**Answers look out of date** — aggregates are cached for up to 10 minutes. Ask
about a specific call or lead for live data.

## Privacy

Aura never receives your Claude or ChatGPT conversations. The connector only
answers the specific questions your assistant asks it, using your own
credentials, scoped to your own workspace.

Lead notes, booking-form answers and AI-extracted quotes are written by third
parties. They are returned to your assistant clearly marked as untrusted
content, so it treats them as data rather than as instructions.