Using the MCP Server

You’ve connected a client — this page covers what you can do with it: which tools appear, how to curate them with custom MCP servers, and what to ask.

What the server exposes

The default server at https://<your-tenant>.alationcloud.com/ai/mcp/ exposes everything you’re entitled to:

Source	What appears	Who sees it
Built-in tools	Catalog search, context retrieval, SQL execution, chart generation, lineage, data quality, and more	Everyone connected to the tenant
Built-in agents	Out-of-the-box agents surfaced as callable tools	Everyone
Your published agents	Any agent with MCP enabled on its Agent access tab	Everyone, if the agent is published; owner and admins only while it is a draft

Because every request carries your identity, two users connecting to the same server can see different tool lists and get different results from the same tool. See identity and permissions for why.

The full catalog of built-in tools is in the tools reference, and the built-in agents in the agents reference. To see your live list, open your client’s tool panel (most clients have one), or simply ask: “What Alation tools do you have available?”

Curate with custom MCP servers

The default server exposes a lot, and agents work best with the smallest tool set their job needs. Instead of disabling tools client by client, create a custom MCP server: a named server that exposes only the tools you pick, at its own URL.

Create the server

From an agent’s Agent access tab, click Add to MCP servers and then New MCP server — or open the MCP servers list in Agent Studio. Give it a name and a description that tells the model what the server is for.
Pick the tools

Click Add tool and select just the tools the use case needs — built-in tools, and agents published as tools.
Configure tool parameters (optional)

Each tool entry has its own parameter settings, with the same Fixed / User input / Agent determined scopes used everywhere in Agent Studio. Parameters you fix are hidden from the connecting model entirely and injected at execution, so a custom server can pin tools to a domain, a data product, or a filter.
Connect your client to the new server

Each custom server has its own URL (/ai/mcp/<server-id>/), shown in the UI. Point your MCP client at that URL instead of the default /ai/mcp/.

Publish an agent to MCP

To make a custom agent callable over MCP — by MCP clients or by other agents:

Open the agent and switch to its Agent access tab.
Turn on the MCP toggle.
Click Add to MCP servers and pick which custom servers should carry it (it is always available on the default server once published).

The agent appears as a tool to everyone once it is published; while it is a draft, only you and admin-tier roles see it.

Example prompts

Discovery and context:

“Find sales-related tables in the catalog.”
“What does the fct_orders table contain, and who stewards it?”
“What’s our approved definition of customer churn?”

Data questions (require a data product and warehouse credentials):

“Using the Sales data product, what were the top 10 products by revenue last quarter?”
“Chart weekly signups from the marketing data product for the past 90 days.”

Governance and lineage:

“Which dashboards report on customer churn?”
“Trace the lineage of dim_customer and summarize the upstream sources.”
“List tables in the raw schema with no description.”

Calling a published agent:

“Use the inventory_checker tool to check stock levels for the EMEA region” — where inventory_checker is an agent you published to the server.

Hosted vs. local servers

Agent Studio supports two ways to use MCP servers: hosted MCP servers and local MCP servers. Both expose Agent Studio capabilities over MCP, but they differ in where the server runs and what it can access.

Hosted MCP servers

Most customers use hosted MCP servers by default. In this mode, the MCP servers are hosted by Alation in our cloud.

Hosted MCP servers can expose:

Base tools
Base (out-of-the-box) agents
Custom agents you build in Agent Studio (including those created in the UI)

This is the best option when you want the full Agent Studio experience and the ability to build, publish, and use custom agents outside of Alation.

Local MCP servers

If you are using the Python SDK, you can also run a local MCP server, meaning you host the MCP server on your own machine using the SDK.

By default, local MCP servers provide access to:

Base tools (data products, data dictionary, custom fields, data sources)
Base (out-of-the-box) agents (catalog context search, query flow, SQL query)

Custom agents you build in the Agent Studio UI always run in Alation, but a local MCP server can expose them too: enable the custom_agent tool (it is off by default) and the server proxies requests to your agent by its ID.

Setup instructions for running a local MCP server are available here.

At a glance

Dimension	Hosted MCP server	Local MCP server (SDK)
Where it runs	Managed by Alation in the cloud	Your machine or infrastructure
Recommended for	Most customers (the default)	Development and SDK-based workflows
Built-in tools	Yes	Yes (configurable subset)
Out-of-the-box agents	Yes	Yes
Custom agents (UI-built)	Yes, after publishing	Via the `custom_agent` tool (off by default); the agent still runs in Alation
Custom MCP servers (curated tool sets)	Yes	Configure the tool set via flags instead
Infrastructure required	None	Self-managed

Good to know

Tool approvals need the web UI. Approval-gated tools can’t be approved from an MCP client; complete the approval in Agent Studio.
Tool calls are metered. Each call to a built-in tool counts as an action for usage purposes, whether it comes from MCP, Slack, or the web UI.
Timeouts apply. Tool calls time out at 30 seconds (configurable to 60 per tool), and some clients impose their own response limits — see known limitations.