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

# MCP Servers

> Connect hosted MCP servers, discover tools, configure caching and approval behavior, and use them safely in Assistant procedures.

Use **Connections → MCP Servers** to connect hosted Model Context Protocol servers to AI Assistant.

MCP lets a server advertise tools with structured input schemas so RAUM can discover and call them during a support workflow.

## Hosted servers only

RAUM is a hosted SaaS platform, so the MCP server must be reachable from the RAUM environment.

> [!WARNING]
>
> Local **stdio** MCP servers are not supported by the hosted RAUM product. Use a hosted remote MCP endpoint instead.

The current connection model uses a server URL and remote HTTP/SSE-style transport.

## Connect a server

Configure the server with:

* name;
* server URL;
* authentication or shared credential;
* description;
* SOURCE/ACTION category where applicable;
* cache TTL; and
* any required headers/settings.

Then use the discovery/test flow so RAUM can read the server's tool list and input schemas.

## Tool-level configuration

Discovered MCP tools can carry their own configuration, including behavior such as:

* tool-specific cache TTL;
* safe-to-simulate state; and
* automatic approval for actions where permitted.

Treat each discovered tool according to its actual effect. One MCP server can expose a harmless lookup next to a tool that deletes something expensive, because protocols are neutral and humans remain committed to adventure.

## SOURCE versus ACTION

Read-only MCP capabilities belong on the SOURCE side of the workflow.

Mutating tools belong on the ACTION side and should retain the applicable approval controls.

## Caching

RAUM can cache discovered tool metadata and tool results according to configured TTL behavior.

Use conservative caching for live account/order data and longer caching only when the source is genuinely stable.

## Use MCP tools in procedures

The procedure editor can reference configured resources so a RUNBOOK can explicitly retrieve or act through the MCP server.

This makes the external dependency visible in the support policy rather than hiding tool behavior inside a generic prompt.

## Simulation

Simulation retrieves SOURCE information so the test has realistic context. ACTION behavior is dry-run/intercepted rather than executed as a real mutation.

Test each mutating MCP tool in Simulation before allowing it into autonomous Assistant handling.

> **[Connections overview](https://docs.raum.am/configuration/integrations.md)**
>
> Understand SOURCE/ACTION safety across API and MCP tools.

> **[Procedures](https://docs.raum.am/configuration/procedures.md)**
>
> Reference tools explicitly from operational runbooks.


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