MCP: connecting the agent to your own services

What Model Context Protocol is, what the built-in ToolsXpo server gives you, and when to add your own.

Updated September 2, 2026

Model Context Protocol (MCP) is an open standard for giving an agent access to tools and data it would not otherwise have. An MCP server exposes capabilities; the agent calls them like any other tool.

The built-in server

toolsxpo-mcp is configured for you when you sign in. It is how the agent reaches the ToolsXpo catalogue — 230+ tools for formatting, conversion, validation, extraction and more.

You cannot delete it, and that is deliberate: it is provisioned on your behalf, and removing it would leave the agent quietly less capable with no error and no obvious way to get it back. The toggle beside it disables it, which is what someone who does not want it actually needs.

Adding your own

MCP → Add Remote Server for a hosted server, or edit the configuration directly for a local one.

Configuration lives at ~/.toolsxpo/data/settings/toolsxpo_mcp_settings.json and is shared by every ToolsXpo app — configure a server once and the editor, CLI and desktop app all have it.

When it is worth it

MCP earns its keep when the agent needs something outside your repository:

your issue tracker, so it can read the ticket it is implementing
your database schema, so it stops guessing at column names
your deployment platform, so it can check what is actually running
an internal API only your team has

If everything the agent needs is in the repository, you probably do not need a server.

What it costs

Every connected server's tool definitions are sent with each request. A server with fifty tools is fifty tool schemas in your context on every turn, whether or not any of them get used.

Connect what you need. Disable what you are not using this week — the toggle is there for exactly this.

Display

Settings → Tools & MCP controls how responses render: rich cards, markdown, or raw text, and whether they are collapsed by default. Some servers return large payloads meant for the agent rather than for you; collapsing those makes a transcript readable again.