# Agentic docs

> Restless isn't just agent-friendly docs. It's the only documentation platform built AI-first. There's a UI for humans, but the docs are shaped to make agents successful.

Source: https://www.restless.ai/agentic-docs

## Features

- A toggle at the top lets humans see exactly what their agents see.
- Your docs aren't for reading; they're for making sure AI can build on your API.
- Agents get more than static docs; they get realtime API data so they can debug what's going on.

## llms.txt

The llms.txt Restless publishes for agents:

```text
# Restless – llms.txt

> You are reading the version of this page built for agents.
> Humans get the UI. You get the structure.

## What Restless does

Turns your API docs into something an agent can act on.  Same source of truth, served in the shape a model parses best.

## Endpoints you can reach

GET   /docs/agent/{resource}          → markdown spec, required + optional fields
GET   /docs/agent/{resource}/schema   → typed params, enums, formats
POST  /mcp                            → live tool calls, realtime API data

## invoices.create

required: amount (integer, cents), currency (ISO-4217)
optional: memo (string, ≤500), due_date (ISO-8601)
returns:  invoice object

## Content negotiation

Send Accept: text/markdown, or crawl as a known AI agent, and every page answers in markdown. No setup.

## Human and agent views

Every reference has two shapes from one source.  Humans get the UI. Agents get the structure.

## Recovery

When a call fails, the response carries a recovery field telling the agent exactly what to send instead.

## Realtime, not static

Endpoints, examples, and responses stream from live traffic, so what an agent reads is never out of date.
```

## Built to be read by agents

Restless generates docs and an MCP server that agents parse cleanly, so they get it right on the first call.

### Agent-Friendly Docs

A public benchmark for how cleanly agents can read your docs. Restless scores a perfect 100.

### MCP Debugger

Checks every tool, schema, and response your MCP server exposes. Restless passes every check.

## Everything we do for agents

- **Markdown for every page** — Add .md to any page URL
- **llms.txt entry point** — The index agents start from
- **Automatic markdown for agents** — AI clients and crawlers get markdown, no setup
- **Human and agent views** — Toggle for humans and agents
- **MCP server** — Access directly where the work is happening
- **Use cases as tools** — Each guide is also a callable tool
- **Auto-generated skills** — Agents onboard themselves, no custom prompts
- **Agent recovery** — Fix-it guidance added to failed responses
- **Workbench over MCP** — Agents debug their own failed calls
- **Agent-native onboarding** — Get an agent to its first call, no docs required

## Agent-first all the way down

Restless is built for agents. Everything is aimed at making them successful and able to figure out how to move forward on your platform.

### Human and agent views

The same reference, two shapes. Humans get the UI. Agents get the structure. One toggle flips between them.

```text
## POST /v1/invoices
required: amount, currency
optional: memo, due_date
returns: invoice object
```

### Agent-shaped markdown

The llms.txt acts as an entry point for agents. Add .md to any page URL and you get it as markdown.

### Served automatically

AI crawlers, and any request that sends Accept: text/markdown, get the markdown version. Nothing to configure.

### Backed by real API data

The docs show actual data from your API, so an agent can see what a call returns and fix its own failed request.

### The same docs power an MCP server

Everything the docs describe is exposed as tools your users' agents can call directly, without reading a page first.

## Related features

- [MCP Server](https://www.restless.ai/mcp-server) — Give agents native access to your API, no docs required.
- [Agent Recovery](https://www.restless.ai/agent-recovery) — Fix-it guidance injected into the response, so failed calls get back on track.
- [AI Traffic](https://www.restless.ai/ai-traffic) — See how agents use your API, split out from developer traffic.

## Setup

Setup is a breeze. Restless is tightly integrated with your API, so we install a small SDK. Don't worry, we walk you through every step of the way.

Run this in your API's repo:

```sh
npx restless init
```

Or hand the setup to your coding agent with this prompt:

```text
Set up Restless in this repo by running `npx restless init` and following the steps.
```

To try it on a demo API instead:

```sh
git clone git@github.com:restlesshq/demo.git && cd demo && npx restless init
```

Nothing is uploaded without your permission.

The CLI runs four steps:

1. **Map out APIs.** Locate your APIs, generate an OpenAPI (OAS) file, and write it to `.restless/`. An OAS file is the shape of your API: every endpoint, parameter, and response. Later steps use it to install the right adapter and wire up the middleware exactly. The scan points Claude Code, running locally on your machine, at your codebase to find your routes. Commit the `.restless/` folder along with your code; it's meant to live there.
2. **Install SDK.** The CLI detects the project runtime, selects its adapter, installs the Restless SDK and wires the request middleware.
3. **Test your setup.** Local traffic capture starts, and the CLI confirms requests are reaching Restless.
4. **Set up account.** Log in to claim your project on Restless.

Privacy: scanning runs entirely on your machine via your own Claude Code install. We don't see a single line of your code, and nothing gets sent to our servers at that step.

## More from Restless

- [Restless](https://www.restless.ai/index.md): Restless makes your API self-improving. Every API call, support question, and error leaves it better for the next request.
- [Features](https://www.restless.ai/features.md): A new surface for your API: a public, LLM-first hub that gets every caller started and keeps them moving.
- [Documentation](https://www.restless.ai/documentation.md): Restless generates your API documentation from your endpoints and the tools your team already uses, and keeps it current every time your API changes.
- [Getting Started](https://www.restless.ai/getting-started.md): Onboarding is the most important part of your developer experience. Restless makes it easy for your developers to get their agent set up and ready to build.
- [Workbench](https://www.restless.ai/workbench.md): The Workbench sits at the bottom of your docs and gives developers a view into everything that's happening: view and filter logs, see what went wrong, and fix issues.
- [Agent recovery](https://www.restless.ai/agent-recovery.md): When an agent hits an error on your API, it usually gives up and you never hear about it. Restless catches the failed call and tells the agent how to fix its request and retry.
- [Human support](https://www.restless.ai/human-support.md): When a developer hits something AI can't fix, a human can step in and answer the question. Every answer becomes part of the context, and makes the next question easier!
- [API Logs](https://www.restless.ai/logs.md): Your API is now how your users use your product. Restless brings every request into one place, so you can see which endpoints get hit, which calls fail, and where people give up.
- [MCP server](https://www.restless.ai/mcp-server.md): Give agents the ability to do more than just write code. Give them access to do everything they need to build on your product, such as reading docs, debugging API calls and updating settings.
- [AI Traffic](https://www.restless.ai/ai-traffic.md): Your API and docs get a lot of AI traffic now, and most of it is invisible. Restless shows you what AI is actually looking for, broken down by who is asking.
- [Self-updating docs](https://www.restless.ai/self-updating-docs.md): Restless reads resolved support conversations, pulls out the part that applies to everyone, and drops it in an inbox to review. Approve it and it becomes context the next person finds on their own.
- [Solicit feedback](https://www.restless.ai/solicit-feedback.md): When an agent hits something confusing in your API, it usually just gives up. Restless adds a send_feedback tool to your MCP server, so the agent can tell you what tripped it up.
- [Sign up](https://app.restless.ai/signup)
