Tutorial · Tools & MCP
How to connect an MCP server to your agent
This tutorial shows you how to give a VegaDūta agent real tools by connecting an MCP (Model Context Protocol) server. You either pick a server from the official MCP registry — browsable and searchable straight from the console — or register your own STDIO or remote server. Once connected, the server's tools appear on your agent, sitting behind guardrails and optional human approval.
Security is built into the flow rather than bolted on: credentials are encrypted at rest, STDIO servers run sandboxed against a command allowlist, remote connections pass SSRF guards, and some registrations go through an admin review before they activate.
Step by step
Open the MCP registry screen
In the console, open the MCP registry. This is the one place for both directions: browsing servers others have published and registering servers of your own.
Browse the official registry or go custom
Search the official MCP registry — the catalog is cached locally, so browsing is fast. If the server you need is not listed, or it is your own, choose 'register custom server' instead.
Enter the server details
For a STDIO server, enter the command and its arguments. For a remote server, enter the URL — SSE and streamable HTTP are both supported — plus any credentials it needs. Credentials are encrypted at rest and sent as headers to remote servers; they are never exposed to the agent as plain text.
Submit for review where required
Some registrations — depending on your workspace's policy — go to an admin for review before they activate. STDIO servers additionally run sandboxed and must match a command allowlist; remote URLs pass SSRF guards. If review is required, the connection shows as pending until approved.
Watch the connection go live
Once approved, the connection activates with liveness tracking, so you can see at a glance whether the server is up. A failing server shows its status here rather than failing silently mid-conversation.
Use the tools from your agent
The server's tools now appear on your agent. Every call goes through guardrails, and you can require human approval for specific tools, so the agent asks before doing anything consequential. Test a tool call in the playground before going live.
Going further
MCP is how a VegaDūta agent stops being a chatbot and starts doing work.
- Home Assistant is a first-party connector — connect it to give an agent control of a real home or office.
- Combine MCP tools with workflow approval nodes for actions that must always be signed off.
- Register internal servers to expose your own systems to agents without opening them to the internet.
Frequently asked questions
What is an MCP server and why would my agent need one?
MCP (Model Context Protocol) is an open standard for giving AI agents tools — actions they can take and data they can read. Connecting an MCP server to a VegaDūta agent gives it those tools, so it can act on systems like calendars, databases, or smart-home devices instead of only answering questions.
Are my MCP server credentials stored safely?
Credentials are encrypted at rest and, for remote servers, sent as request headers when the platform calls the server. They are never handed to the model as plain text. You can rotate or remove them from the registry screen at any time.
Can I connect a local STDIO MCP server, or only remote ones?
Both. STDIO servers are registered with a command and arguments and run sandboxed against a command allowlist; remote servers connect over SSE or streamable HTTP behind SSRF guards. Both kinds get liveness tracking once connected.
How do I stop an MCP tool from being used without permission?
Every MCP tool call already passes through the agent's guardrails, and you can additionally require human approval for specific tools. The agent then pauses and asks before executing, which is the recommended setup for anything that changes state.
See it working in two minutes
The sandbox provisions a real tenant — describe an agent in one sentence and test it, no account, no card. Or browse ~90 industry workflow recipes to see what teams build.
