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

# Quickstart: AI Agents

Connect your AI agent to Membrane and give it the ability to interact with any external app — connect accounts, run actions, read data, and subscribe to events.

Pick the option that matches your agent:

| Your agent                                                          | Use                                            |
| ------------------------------------------------------------------- | ---------------------------------------------- |
| Coding agent with CLI access (Claude Code, Cursor, Windsurf, Codex) | [Agent Skills](#agent-skills)                  |
| MCP-compatible agent (Claude Desktop, ChatGPT, etc.)                | [MCP Server](#mcp-server)                      |
| Custom agent or in-app AI assistant                                 | [Custom / In-app Agent](#custom--in-app-agent) |

## Agent Skills

If your agent supports skills and you can run commands in the terminal, this is the fastest path.

```bash theme={null}
npx skills add membranehq/agent-skills --skill integrate-anything
```

This gives your agent Membrane's integration tools. It sets up the MCP server and credentials automatically. Prompt your agent:

> "Connect to HubSpot and create an action that syncs new contacts to our database."

See [Agent Skills](/docs/ways-to-use-membrane/agent-skills) for details.

## MCP Server

If your agent supports MCP but not skills, add Membrane's hosted MCP server:

```json theme={null}
{
  "mcpServers": {
    "membrane": {
      "type": "url",
      "url": "https://api.getmembrane.com/mcp/integrate-anything"
    }
  }
}
```

The hosted server handles authentication via OAuth — your agent will prompt you to log in when it first connects.

See [MCP Server](/docs/ways-to-use-membrane/mcp) for platform-specific config file locations (Claude Desktop, Cursor, etc.).

## Custom / In-app Agent

If you're building your own agent or embedding an AI assistant in your product:

* **Custom standalone agent** — see the [agent-skills repository](https://github.com/membranehq/agent-skills) for ready-to-run examples with OpenAI SDK, Vercel AI SDK, LangChain, and OpenCode adapters.
* **In-app AI assistant** (multi-tenant, acts on behalf of each user) — see [In-app AI Agent](/docs/getting-started/product-integrations/in-app-ai-agent) for the complete guide with per-user authentication and scoped connections.