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

# Agent discovery files Dedica serves

> Dedica serves ard.json, ai-catalog.json, an MCP server card, an agent skill, openapi.json and an RFC 9727 api-catalog so agents can find what your site offers.

Discovery files tell agents what your site offers them: where your `llms.txt` is, your MCP server, `/ask` and your API. Dedica builds them from what it already serves for your site and serves them on your domain while your site is in [autopilot](https://docs.getdedica.com/guides/serving-modes.md). Nothing is stored separately, so they stay in step with what you approve.

| Path | Standard | Served while |
| - | - | - |
| `/.well-known/ard.json` | ARD v0.91 | Any file below exists |
| `/.well-known/ai-catalog.json` | AI Catalog 1.0 | Same |
| `/.well-known/mcp/server-card.json` and `/mcp/server-card` | MCP server card | You have an approved agent tool |
| `/.well-known/agent-card.json` | A2A Agent Card v0.3.0 | You have an approved agent tool |
| `/.well-known/agent-skills/index.json` | Agent Skills discovery v0.2.0 | `llms.txt` is served, or you have an approved agent tool |
| `/.well-known/agent-skills/<your domain>/SKILL.md` | Agent Skills | `llms.txt` is served |
| `/.well-known/agent-skills/<tool name>/SKILL.md` | Agent Skills | That agent tool is approved and on |
| `/openapi.json` | OpenAPI 3.1 | You have an approved HTTP agent tool |
| `/.well-known/api-catalog` | RFC 9727 | Same |
| `/schemamap.xml` | NLWeb Schema Map | A page Dedica serves has schema.org structured data (JSON-LD) |

`llms.txt` also lists the files served right now in a "Machine-readable files" section.

## Your own file always wins

If your site already serves one of these files, agents get yours. The middleware asks your site first and only serves Dedica's file when your site has none.

## Agent skills

The skills index lists one skill for your whole site (when and how to use it) and one per agent tool. A tool's skill says what the tool does, lists its arguments and shows how to call it through your [MCP server](https://docs.getdedica.com/serving/mcp-and-ask.md) or, for HTTP tools, as a plain request. Skills for tools that change something tell the agent to confirm with the person first. A tool gets its skill name from its own name, `get_report` becomes `get-report`.

## A2A agent card

The agent card lists your agent tools as A2A skills. Its endpoint is your MCP server, where each skill is the tool of the same name. Dedica doesn't run an A2A server, so a client that only speaks A2A's own methods can't call it.

## Schema map

`/schemamap.xml` lists the pages Dedica serves agent versions of, once for each schema.org type in the page's own JSON-LD, so agents can fetch the structured data directly. Pages without JSON-LD aren't listed. To point crawlers at it, add `schemamap: https://<your domain>/schemamap.xml` to your `robots.txt`.

## OpenAPI from your tools

`/openapi.json` has one operation per approved HTTP agent tool. Link tools aren't included because they aren't requests. Because tool calls are anonymous, the spec declares no security scheme. Every endpoint in it is already callable through your [MCP server](https://docs.getdedica.com/serving/mcp-and-ask.md), so the spec exposes nothing new.


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