MCP Connections#

Overview#

The Model Context Protocol (MCP) is an open standard for connecting AI systems to external tools. An MCP connection brings tools from an external service — an issue tracker, a knowledge base, an internal API — into Roboto’s AI, so it can consult those systems mid-conversation alongside its built-in tools.

Note

MCP also works in the other direction: the Roboto MCP Server brings Roboto’s own tools to external AI apps like Claude Code, Claude Desktop, Codex, Cursor, and VS Code.

Key properties:

  • Roboto makes MCP servers available to your organization. To connect a custom MCP server, contact support@roboto.ai.

  • Connections are per user: each person authorizes the external service with their own account, and the AI acts with that person’s access — never someone else’s.

  • Tools are discovered lazily: the AI fetches a server’s tool list only when a conversation calls for it.

  • Tool availability is curated by an allowlist. By default, if any of a server’s tools is annotated read-only, only tools explicitly marked read-only are enabled — the AI consults external systems but doesn’t write to them.

  • Org admins can add context for each server: a short description of how your organization uses it. The AI sees this note in every conversation, so it knows when to reach for the server and uses it the way your team expects.

Connecting to a server#

In the web app, go to Settings → AI → Personal MCP Connections. The page lists the servers available to your organization; click Connect and sign in to the external service in the popup. Disconnect at any time from the same page.

Each user connects individually. A teammate’s connection never grants you access to the external service, and yours never grants them access.

If your authorization expires, the server shows as Expired — click Reconnect to re-authorize.

Note

If you’re an org admin, Connect stays disabled for a server until you’ve added its AI context (see Administration below).

How the AI uses external tools#

The AI discovers external tools lazily. It always knows which MCP servers you’ve connected; when a conversation calls for one, it fetches that server’s tool list mid-conversation, and the enabled tools become available for the rest of the thread — used exactly like built-in tools, with results appearing in the conversation.

You can also point the AI at a server explicitly: “check GitHub for open issues about this firmware version”.

Administration#

  • Availability: Roboto registers MCP servers and scopes them to specific organizations (or makes them broadly available). Contact support@roboto.ai to request a custom server.

  • Tool allowlist: a per-server allowlist controls which of a server’s tools the AI may use. On first discovery the allowlist is seeded automatically: if any of the server’s tools carries MCP’s read-only annotation, only tools explicitly marked read-only are enabled — everything else, including unannotated tools, is blocked. Servers that don’t use the annotation at all start with every tool enabled. Roboto manages the allowlist; contact support@roboto.ai to change which tools are enabled.

  • Org context: org admins can attach a short plaintext note to each server describing how your organization uses it. The AI sees this note in every conversation — even for servers a user hasn’t connected yet — so it can suggest connecting when relevant.

See the Connect External Tools via MCP user guide for a step-by-step walkthrough.

On this page