MCP Servers

MCP Servers is the RealTimeX settings area for connecting external tool servers that follow the Model Context Protocol.

Open it from Settings > Agents > MCP Servers.

What this page is for

Use MCP when you want agents to call tools outside the core RealTimeX feature set.

Typical examples include:

external APIs and SaaS actions
developer tools and databases
filesystem or command wrappers
hosted MCP services that require account linking
local MCP processes that run on your own machine or deployment

In the current product, the main user-facing value of MCP is tool access inside agent workflows and agent chat sessions.

Local vs remote

The current MCP screen is split into two tabs.

Remote MCP

Remote MCP is for hosted MCP servers that RealTimeX can discover and connect to over the network.

Use this tab when:

the MCP provider is already hosted for you
the server needs account linking or API-key auth
you want to browse available hosted servers and connect them in-app

Remote servers are typically connected rather than authored. After you open a server, RealTimeX can prompt for:

no-auth connection
API key
OAuth 2 authorization

After a remote server is linked, you can review its available tools and decide which functions to enable.

Local MCP

Local MCP is for MCP servers that run in your own environment.

Use this tab when:

you want to run an MCP server on your own machine
you want to run an MCP server inside your self-hosted deployment
you need custom command, URL, or transport configuration
you want to manage a server template or imported config directly

In the current app, local servers are first-class records inside RealTimeX. You can add, configure, enable, disable, and delete them from the UI.

Typical workflow

Open Settings > Agents > MCP Servers.
Choose Remote MCP or Local MCP.
Connect or configure the server.
Review the discovered tool list.
Enable only the tools you want exposed.
Assign the server where your agents can use it.

The last step matters. Connecting a server does not automatically make it active in every workspace or agent configuration.

Tool access controls

Both local and remote MCP flows expose per-tool visibility controls in the current UI.

Use them when:

a server exposes many tools but you only want a safe subset
you want to reduce agent confusion or accidental tool calls
you are testing one tool before exposing the rest

RealTimeX shows tool availability only after the server is in a usable state:

local servers must be enabled and configured
remote servers must be connected and have a valid server configuration

Where MCP servers are actually used

After a server is connected, you still need to expose it to the relevant agent context.

The main place to do that is the workspace agent configuration flow, where RealTimeX lets you choose which remote and local MCP servers are available to that workspace or agent setup.

For the agent side of that workflow, see Using AI Agents and Agent Setup.

Local MCP setup in RealTimeX

The current local-server flow is UI-first.

RealTimeX can currently support:

built-in or suggested server templates
custom local server entries
configuration import or raw config editing
placeholder-driven setup fields
connection testing before saving
per-server enable or disable state
per-tool enable or disable state

This is a much better fit for end users than manually editing JSON files unless you already know exactly what server definition you want to load.

Remote MCP setup in RealTimeX

The current remote-server flow is connection-first.

RealTimeX can currently support:

browsing available remote servers
connecting with the auth method required by the provider
updating linked access later
reviewing the available tool list
enabling only selected functions

Some remote providers use OAuth and return you to the app after authorization. Others ask for an API key or require no auth at all.

Deployment notes

The operational meaning of local depends on where RealTimeX is running.

On Desktop, a local MCP server runs on your device.
On self-hosted or containerized deployments, a local MCP server runs in that deployment environment.
Remote MCP is hosted elsewhere and connected over the network in either case.

See the deployment-specific guides:

Safety notes

Only connect MCP servers you trust.
Prefer enabling only the tools you actually need.
Treat API keys and OAuth-connected servers as production integrations, not toy examples.
Review what a tool can read, write, or trigger before exposing it to an agent.

Related guides

Credentials MCP on Desktop