# Written by agents. Read by agents.

Manicule’s agents write and test your docs. Margin serves them to every agent that asks.

[Deploy free on Manicule](https://app.manicule.com/onboard?product=margin) · [View on GitHub](https://github.com/maniculehq/margin)

- **Read by agents:** Free in every Margin site
- **Written by agents (beta):** Manicule’s writing agents

## Every page, two formats.

Add .md to any URL for Markdown, or .pdf for a download.

## One page, two readers.

Write for agents without cluttering the page people read.

### People see the page

The tour image shows. The agent note stays out of the HTML.

### Agents get the Markdown

The note is in. The tour image is gone.

Components: `<Visibility>`, `<Agent>`, `<Human>`

## Files agents look for.

Generated from each deploy. Nothing to write or maintain.

- `/llms.txt`: Every page, one link each
- `/llms-full.txt`: Every page in one file
- `/sitemap.md`: The navigation as Markdown
- `/AGENTS.md`: Instructions for coding agents
- `/skill.md`: Your docs as a skill
- `/changes.md`: What changed in the docs
- `/.well-known/llms-manifest.json`: A machine-readable corpus manifest
- `/.well-known/mcp.json`: Where the MCP server lives
- `/.well-known/agent-skills/index.json`: Every published skill, listed
- `/.well-known/api-catalog`: Your OpenAPI specs, when present

### What is llms.txt?

llms.txt is a plain-text file at the root of a site, written for language models. It lists every page with a link and a one-line summary, so an agent can find the right page without crawling the HTML. Its companion, llms-full.txt, holds the full text of every page in one file.

## A built-in MCP server.

Add one URL to any MCP client. On by default, on every site.

Tools:

- `search_docs(query, limit?, url?, locale?, version?)`
- `read_page(url)`
- `list_docs(url?, prefix?, limit?, cursor?)`
- `get_navigation(url?)`
- `get_code_examples(url, language?)`
- `query_docs_filesystem(command, url?, locale?, version?)`
- `submit_feedback(url, kind, message, suggestedFix?)`

Prompts:

- `answer_from_docs(question)`
- `implement_with_docs(task)`

Client config:

```json
{
  "mcpServers": {
    "acme": { "url": "https://docs.acme.dev/api/mcp" }
  }
}
```

## Wherever agents work.

Hand any page to an AI tool, or give browser agents tools to use the site.

### Copy page menu

Copy page, View as Markdown, Open in ChatGPT, Open in Claude, Connect to Cursor, Open in T3 Chat

### Add more in docs.json

Download PDF, Connect MCP server, Install MCP in VS Code, Open in Perplexity, Open in Grok, Open in AI Studio

### Browser tools (WebMCP)

Experimental, off by default. Tools: `search_docs`, `read_page`, `get_current_page_context`, `get_site_overview`, `read_skill`, `open_documentation`

## See what agents read.

Hosted sites chart human and AI-agent views of your docs, day by day.

Then see which AI answers cite them. [Track visibility in Manicule](https://manicule.com/agent-visibility)

## Written by agents (beta)

Manicule’s writing agents draft and update your pages from what your team already knows.

### Grounded in your context

It reads your team’s documents before it writes.

### Written to be used

Outcome first, with working code near the top.

### Better every run

Confirmed facts carry over to the next page.

## Tested by a coding agent (beta)

It builds what each page teaches against your real API, using only the page.

## You approve every change.

### Review a page in chat

Ask for changes in plain words, then read the new version.

### Edit many pages at once

One instruction, applied across every page you pick.

### Every change is a version

Earlier versions stay, so you can compare and go back.

## Try the writing agents (beta)

Book a call and we’ll set them up on your docs. [Book a call](https://savvycal.com/manicule/intro)

## Publish your docs today.

Free on Manicule in a few minutes, or on your own servers whenever you like.

[Deploy free on Manicule](https://app.manicule.com/onboard?product=margin) · [View on GitHub](https://github.com/maniculehq/margin)