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

# Custom MCP servers

> Bring your own tools into Swan by connecting any MCP-compatible server.

If a tool you use isn't in Swan's [integrations catalog](built-in-integrations), you can connect it directly through MCP (Model Context Protocol) — an open standard many tools now support for exposing their capabilities to AI agents. This page covers Swan as the MCP *client*, consuming your server's tools. The other direction exists too — Swan is also an MCP server your own agents can connect to and talk to: see [connect your agent to Swan](connect-your-agent-to-swan).

## Connecting a server

From [Tools](https://agent.getswan.com/tools), click "Add MCP Server" and give Swan its URL and how to authenticate — either an HTTP header/query-param value (like an API key) or an OAuth2 flow, depending on what the server requires. There's no support for importing a JSON config file (like a `claude_desktop_config.json`) — you configure the connection directly through this URL-and-auth form.

## Presets

A handful of common servers have ready-made presets, so you only need to supply credentials rather than the raw URL — today: Monday.com, Close CRM, SimilarWeb, Mixpanel, HeyReach, Outreach, Gamma, and Zapier. These show up as cards alongside the rest of the [integrations catalog](built-in-integrations) in [Tools](https://agent.getswan.com/tools).

Presets use whatever auth method that server expects — most are OAuth, a couple (SimilarWeb, HeyReach, Zapier) use an API key or token passed as a header or query parameter instead.

You're not limited to these — any SSE- or HTTP-based MCP server can be added as a general custom connection, preset or not.

## How connected tools show up

Once connected, the server's tools appear to the agent as their own toolkit, alongside your built-in and app-catalog toolkits. You (or Swan) can reference them in triggers, templates, and conversations the same way as any other tool. Open the server's card in [Tools](https://agent.getswan.com/tools) to see its discovered tools, re-sync them after the server's tool list changes, or edit its name, headers, and custom instructions.

## Cost

Each call to a tool from a connected MCP server costs 1 credit by default, the same baseline as a HubSpot or app-catalog write. See [credits](../administration/credits) for the full picture of what costs what.

## Security

Tool output from your MCP server flows directly into the agent's context, the same as any other tool result — the agent reads and reasons over whatever the server returns. Only connect servers you trust, and be mindful of what data a server can see or return before wiring it in.

## Common questions

**Can Swan act as an MCP server, not just a client?**
Yes — Swan exposes its own MCP server that any MCP-capable agent can connect to and hand work to. See [connect your agent to Swan](connect-your-agent-to-swan).

**Do I need to know the exact MCP server URL, or can I just pick from a list?**
Either — use a preset if the server you want is listed, or enter any MCP server's URL manually for anything else.

**What if the server requires OAuth?**
Swan walks you through an OAuth flow when you add the server, the same as connecting an app from the catalog.

**Does a custom MCP connection replace the built-in integrations?**
No, it's additive — use MCP for tools Swan doesn't already integrate with natively.

**The tool I want doesn't expose an MCP server either — now what?**
There's one more route: store the tool's API key as a secret and have Swan call its API directly. See [direct API access](direct-api-access).

**Can I limit which of a connected server's tools the agent is allowed to use?**
Yes — open the server's card in [Tools](https://agent.getswan.com/tools) (Capabilities tab) and you'll see every tool it exposes, each with the same enable/disable and approval controls as tools from any built-in integration.