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

# Agent quickstart

> Give a connected assistant a bounded discovery-to-CRM workflow with clear approval gates.

First [connect with OAuth](https://docs.outreign.io/developer-docs/mcp/connect.md) to `https://api.outreign.io/mcp`. The docs URL is not a connector endpoint. A URL pasted into an ordinary chat does not install or authorize the connector.

## Verify the connection

Ask: **“List up to five saved OutReign leads. Do not modify anything or send messages.”**

The client should discover `list_leads` via `tools/list`. Empty results can mean the account has no saved leads; “failed\_to\_load”, authorization errors or a tool `isError` are not an empty successful result. The current source catalog contains 24 tools; availability of a new revision depends on app publication and refreshing the connector.

## Bounded prospecting prompt

```text theme={null}
Use my OutReign connection to find up to 10 founders or sales leaders at
small B2B SaaS, tech or marketing agencies in my selected geography.
Do not include local consumer-service businesses. Do not broaden the
targeting without asking me.

Discover the actual tool schemas first. Use search_prospects for new
prospects, not search_leads, which searches saved CRM records.
Show company, role, fit rationale and returned sources where available.
Do not invent contact details. If an email is returned, check it with
verify_email and preserve unknown/catch_all/role verdicts as unconfirmed.

Show candidates for my approval before saving, enriching or spending
lookup credits. After I select records, create_lead, retain the saved
UUID and use it for enrichment, research, scoring and drafting.
Never enroll sequences, send replies or enable automation without
separate approval of the exact action. Report failures honestly.
```

## What each step proves

| Step | Evidence | Not proof of |
| - | - | - |
| tools/list | Available schemas | Permission or quota for every call |
| search\_prospects | Returned network candidates | Saved CRM records or email validity |
| verify\_email | Mailbox verdict | Consent or guaranteed delivery |
| create\_lead | Returned saved lead UUID | Enrichment completion |
| research\_lead | Brief and cited sources | A buying commitment |
| generate\_email | Draft copy and saved draft\_id in Unified Inbox | Email sent or delivered |
| send\_reply | Queued reply | Delivered message |

## Workspace confirmation

For `workspace_action`, first request the exact edit without `confirmed`. Review the returned `confirmation_required` preview. After the user approves, repeat the same action and arguments with `confirmed: true`. Approval is a client/user responsibility; do not infer consent from a previous unrelated request.

Other outbound tools must also respect user approval; a tool accepting arguments is not proof the user approved it. Pausing/resuming sequences or enabling an agent can change later automated sending.

## Recover safely

Refresh cached tools after publication. Reconnect if authorization expired or was revoked. Inspect MCP `isError` and app-action JSON errors. If the client reports a challenge page/403 rather than a JSON tool error, preserve the attempt time and client name for support. Never share tokens.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.