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

# For agents

> Connect your assistant to Federated Agent Use — resolve users, sign MCP calls, get answers.

## You are here because

You ship an AI assistant (or agent framework) and want to **ask a person's mail node
for answers** without taking their Gmail credentials.

**Primary goal:** connect to the network and complete a first successful tool call
(usually after a phone approval).

## What you will build

```
Your agent  →  relay.fau.run/v1/resolve
            →  TLS to their node hostname
            →  signed POST /mcp  (tools/call)
            →  show match code if approval_required
            →  retry after they allow on their phone
```

You do **not** run their Mac, sync IMAP, or hold an inbox token.

## How to integrate

| Your stack | Path |
| - | - |
| **Python** | Prefer the [Python agent SDK](https://docs.fau.run/developer/agents/python-sdk.md) (`fau-agent`) |
| **Any other language** | Speak protocol `fau/1` yourself — [Connect your agent](https://docs.fau.run/developer/agents/getting-started.md) |

The protocol is the contract. The SDK is the reference client for Python so you do not
re‑implement signing, sessions, or the offline queue.

## Paste into Cursor / Claude / Codex

Copy this prompt (or open it in Cursor). The agent should pull the linked docs before writing code.

Connect an agent to FAU — with docs as context

Help me connect an AI agent to Federated Agent Use (protocol `fau/1`).

Goal: resolve a user on the public relay, open a signed MCP session to their node,
call a tool, and correctly handle `approval_required` (show domain + match code, then retry).

If I am using Python, prefer the fau-agent SDK
([https://docs.fau.run/agents/python-sdk](https://docs.fau.run/agents/python-sdk) and sdk/python in the repo).
Otherwise implement the wire protocol from the docs — do not invent APIs.

Read these docs first (fetch as Markdown if your client supports it):

* [https://docs.fau.run/agents/getting-started](https://docs.fau.run/agents/getting-started)
* [https://docs.fau.run/agents/python-sdk](https://docs.fau.run/agents/python-sdk)
* [https://docs.fau.run/connect/signing](https://docs.fau.run/connect/signing)
* [https://docs.fau.run/connect/transport](https://docs.fau.run/connect/transport)
* [https://docs.fau.run/tools/approvals](https://docs.fau.run/tools/approvals)
* [https://docs.fau.run/reference/checklist](https://docs.fau.run/reference/checklist)

Hard rules from the docs:

* Prefer structured tools and `ask` before `get_excerpt`.
* Treat `approval_required` as normal control flow.
* Always show verified agent domain and 2-digit match code.
* Never request Gmail passwords or inbox OAuth.

Public relay: [https://relay.fau.run](https://relay.fau.run)
Start by outlining the integration steps, then implement against my stack when I specify language/framework.

## Start here

1. **Python?** → [Python agent SDK](https://docs.fau.run/developer/agents/python-sdk.md)
2. **Any language / understand the wire** → [Connect your agent](https://docs.fau.run/developer/agents/getting-started.md)
3. Then deepen as needed:

| Task | Page |
| - | - |
| Resolve email → hostname | [Resolve](https://docs.fau.run/developer/connect/resolve.md) |
| Publish agent keys | [Identity](https://docs.fau.run/developer/connect/identity.md) |
| No public domain yet | [Enroll with a claim](https://docs.fau.run/developer/connect/enroll.md) |
| Sign every `/mcp` request | [Signing](https://docs.fau.run/developer/connect/signing.md) |
| MCP session + TLS | [Transport](https://docs.fau.run/developer/connect/transport.md) |
| Mac is asleep | [Offline queue](https://docs.fau.run/developer/connect/offline-queue.md) |
| Tool catalog | [Tools](https://docs.fau.run/developer/tools/overview.md) |
| Phone prompts | [Approvals](https://docs.fau.run/developer/tools/approvals.md) |
| Ship checklist | [Integration checklist](https://docs.fau.run/developer/reference/checklist.md) |

## Mental model (one sentence)

The relay finds the user's door; your signed tool call knocks; their **phone** unlocks
access; their **node** answers from local mail.

## Hard rules

* Prefer structured tools and `ask` before `get_excerpt`.
* Treat `approval_required` as normal control flow, not a crash.
* Always show the **verified agent domain** and **2-digit match code** to the user.
* Never request Gmail passwords or inbox OAuth for your product.

## Not this section

Running a node, deploying a relay, or building the phone app →
[For protocol developers](https://docs.fau.run/developer/developers.md).


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