Aptos provides first-class support for AI-powered development workflows. Whether you’re using an AI coding assistant or an AI chat tool, you can give it deep knowledge of the Aptos blockchain, Move language, SDKs, and APIs.

## Aptos MCP Server

[](#aptos-mcp-server)

The [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) lets AI coding tools directly call Aptos APIs, query on-chain data, and generate correct Aptos code — without needing to copy-paste documentation.

[Claude Code](/build/ai/aptos-mcp/claude) Set up the Aptos MCP server for Claude Code

[Cursor](/build/ai/aptos-mcp/cursor) Set up the Aptos MCP server for Cursor

[OpenAI Codex](/build/ai/aptos-mcp/codex) Set up the Aptos MCP server for Codex

Any MCP client that can launch a local stdio server — including GitHub Copilot — can use the same `npx -y @aptos-labs/aptos-mcp` command. See the [Aptos MCP overview](/build/ai/aptos-mcp) for prerequisites and the Geomi Bot Key.

## Aptos Agent Skills

[](#aptos-agent-skills)

[Aptos Agent Skills](https://github.com/aptos-labs/aptos-agent-skills) provides pre-built AI skills that give coding assistants deep Aptos expertise — covering Move smart contracts, the TypeScript SDK, and full-stack dApp scaffolding.

```
npx skills add aptos-labs/aptos-agent-skills
```

This command uses the [Skills CLI](https://www.npmjs.com/package/skills), which is invoked via `npx`. If `npx skills` is not available in your environment, ensure you have Node.js and npm installed and refer to the Skills CLI documentation for installation and troubleshooting details.

[Aptos Agent Skills](/build/ai/aptos-agent-skills) Browse available skills for Move development, TypeScript SDK, and project scaffolding

## LLMs.txt Documentation Feeds

[](#llmstxt-documentation-feeds)

The [LLMs.txt standard](https://llmstxt.org/) provides machine-readable documentation feeds that AI tools can ingest to understand the full Aptos documentation. We publish three feeds to suit different context window sizes:

<table><thead><tr><th>Feed</th><th>URL</th><th>Best for</th></tr></thead><tbody><tr><td><strong>llms.txt</strong></td><td><a href="/llms.txt"><code dir="auto">/llms.txt</code></a></td><td>Structured index with page titles, descriptions, and per-page <code dir="auto">.md</code> links</td></tr><tr><td><strong>llms-small.txt</strong></td><td><a href="/llms-small.txt"><code dir="auto">/llms-small.txt</code></a></td><td>Condensed docs for smaller context windows</td></tr><tr><td><strong>llms-full.txt</strong></td><td><a href="/llms-full.txt"><code dir="auto">/llms-full.txt</code></a></td><td>Full documentation — all pages concatenated</td></tr></tbody></table>

Some clients resolve `/.well-known/llms.txt`; production serves the same index via a redirect to [`/llms.txt`](/llms.txt) ([`https://aptos.dev/.well-known/llms.txt`](https://aptos.dev/.well-known/llms.txt)).

The [`llms.txt`](/llms.txt) router also surfaces machine-readable **API** links—[`/aptos-spec.json`](/aptos-spec.json) (OpenAPI 3 JSON for the node REST API) and [`/rest-api`](/rest-api) (HTML reference)—plus **MCP**, Agent Skills, Explorer, GitHub, standards, and Indexer GraphQL `.md` links in one place.

[How to use LLMs.txt with AI tools](/llms-txt) Instructions for Cursor, Claude Code, GitHub Copilot, ChatGPT, Codex, Devin Desktop, and more

## Agent discovery endpoints

[](#agent-discovery-endpoints)

The docs site publishes a handful of standardized discovery documents so coding agents can bootstrap from a single root. Every page also returns a [`Link`](https://www.rfc-editor.org/rfc/rfc8288) response header and matching `<link rel="…">` tags that advertise the same URLs.

<table><thead><tr><th>Endpoint</th><th>Purpose</th></tr></thead><tbody><tr><td><a href="/.well-known/api-catalog"><code dir="auto">/.well-known/api-catalog</code></a></td><td><a href="https://www.rfc-editor.org/rfc/rfc9727">RFC 9727</a> linkset pointing at the REST API spec, documentation, Indexer GraphQL, and the LLMs.txt feeds.</td></tr><tr><td><a href="/.well-known/mcp/server-card.json"><code dir="auto">/.well-known/mcp/server-card.json</code></a></td><td><a href="https://github.com/modelcontextprotocol/modelcontextprotocol/pull/2127">MCP Server Card (SEP-2127)</a> describing the <a href="https://www.npmjs.com/package/@aptos-labs/aptos-mcp"><code dir="auto">@aptos-labs/aptos-mcp</code></a> server.</td></tr><tr><td><a href="/.well-known/agent-skills/index.json"><code dir="auto">/.well-known/agent-skills/index.json</code></a></td><td><a href="https://github.com/cloudflare/agent-skills-discovery-rfc">Agent Skills Discovery</a> index for <a href="https://github.com/aptos-labs/aptos-agent-skills">aptos-agent-skills</a> with per-skill SHA-256 digests.</td></tr><tr><td><a href="/.well-known/oauth-protected-resource"><code dir="auto">/.well-known/oauth-protected-resource</code></a></td><td><a href="https://www.rfc-editor.org/rfc/rfc9728">OAuth 2.0 Protected Resource Metadata (RFC 9728)</a> for <code dir="auto">https://aptos.dev</code>. Lists the resource identifier, the Firebase + Google authorization servers used by the <a href="/network/faucet">Aptos Testnet Faucet</a>, and supported scopes.</td></tr><tr><td><a href="/.well-known/openid-configuration"><code dir="auto">/.well-known/openid-configuration</code></a></td><td><a href="https://openid.net/specs/openid-connect-discovery-1_0.html">OpenID Connect Discovery 1.0</a> document advertising the issuer, authorization/token endpoints, JWKS URI, and supported grant types used by the <a href="/network/faucet">Aptos Testnet Faucet</a> sign-in flow.</td></tr><tr><td><a href="/.well-known/oauth-authorization-server"><code dir="auto">/.well-known/oauth-authorization-server</code></a></td><td><a href="https://www.rfc-editor.org/rfc/rfc8414">OAuth 2.0 Authorization Server Metadata (RFC 8414)</a> mirror of the OIDC discovery document for clients that prefer the pure OAuth 2.0 endpoint name.</td></tr><tr><td><a href="/.well-known/ai-catalog.json"><code dir="auto">/.well-known/ai-catalog.json</code></a></td><td><a href="https://agenticresourcediscovery.org/">ARD</a> / <a href="https://github.com/Agent-Card/ai-catalog">AI Catalog</a> capability manifest listing the MCP Server Card, Agent Skills index, REST OpenAPI spec, and RFC 9727 API catalog.</td></tr><tr><td><a href="/auth.md"><code dir="auto">/auth.md</code></a></td><td><a href="https://workos.com/auth-md">auth.md</a> agent authentication instructions. Documents that <code dir="auto">aptos.dev</code> is public (no agent registration) and that the Testnet Faucet uses human Google OIDC.</td></tr><tr><td><a href="/robots.txt"><code dir="auto">/robots.txt</code></a></td><td>Includes <a href="https://contentsignals.org/">Content Signals</a> (<code dir="auto">ai-train=yes, search=yes, ai-input=yes</code>), an <code dir="auto">Agentmap:</code> pointer to the AI Catalog, and <code dir="auto">Sitemap: https://aptos.dev/sitemap.xml</code>.</td></tr></tbody></table>

The origin-level RFC 9728 document identifies `https://aptos.dev` so clients that require the `resource` identifier to match this host will accept it. Documentation on this host is unauthenticated. The listed authorization servers (Firebase + Google) mint the bearer tokens that the [Aptos Testnet Faucet](/network/faucet) accepts on `Authorization: Bearer …`; the faucet API itself lives at `https://faucet-e2.testnet.aptoslabs.com`. [`/auth.md`](/auth.md) is the prose companion: there is no agent-registration endpoint.

### DNS for AI Discovery (DNS-AID)

[](#dns-for-ai-discovery-dns-aid)

[DNS-AID](https://datatracker.ietf.org/doc/draft-mozleywilliams-dnsop-dnsaid/) records cannot be published from this repository. The `aptos.dev` zone is hosted on Google Cloud DNS (`ns-cloud-c1.googledomains.com` through `ns-cloud-c4.googledomains.com`).

The organization index already exists:

```
_index._agents.aptos.dev. 3600 IN HTTPS 1 aptos.dev. alpn="h2,h3" port=443
```

That record points agents at `https://aptos.dev`, where the MCP Server Card, API catalog, AI Catalog, and other well-known documents live. Do not publish `_a2a._agents` or `_mcp._agents` until Aptos hosts those protocols on HTTPS — the MCP server is stdio (`npx @aptos-labs/aptos-mcp`), not a remote MCP endpoint.

The zone has `DNSKEY` and `RRSIG` records, but validating resolvers still return `AD=0` because there is no `DS` record at the `.dev` parent. Agent scanners that require DNSSEC therefore fail even though the HTTPS record is present. Operators should publish the KSK `DS` at the registrar (or enable DNSSEC in Google Cloud DNS so it submits the `DS` automatically), then confirm:

```
dig +short HTTPS _index._agents.aptos.devdig +dnssec HTTPS _index._agents.aptos.devdig DS aptos.dev
```

To advertise the AI Catalog over DNS as well, add:

```
_catalog._agents.aptos.dev. 3600 IN TXT "url=https://aptos.dev/.well-known/ai-catalog.json"
```

### Markdown content negotiation

[](#markdown-content-negotiation)

Any docs URL that has a `.md` mirror (for example [`/build/sdks`](/build/sdks) → [`/build/sdks.md`](/build/sdks.md)) will also respond to `Accept: text/markdown` with `Content-Type: text/markdown`:

```
curl -H "Accept: text/markdown" https://aptos.dev/build/sdks
```

HTML stays the default for browsers that send the usual `Accept: text/html`.

### WebMCP

[](#webmcp)

Chrome and other browsers that implement the [WebMCP proposal](https://webmachinelearning.github.io/webmcp/) can discover the following tools on every docs page via `navigator.modelContext`: `aptos-docs.search`, `aptos-docs.open-doc`, `aptos-docs.fetch-doc-markdown`, and `aptos-docs.list-llms-feeds`.