What Is Atlan MCP and How Does It Work?

Atlan MCP is a server that gives AI agents data-catalog and governance metadata from Atlan. It works in Claude Code, Cursor, Windsurf, VS Code with Copilot and other MCP clients, and installs with mcpizy install atlan. As of November 2026, the CheckMCP audit of its repository scores it 87/100 (grade B). Because it exposes metadata about your data estate, access control deserves a review before you connect it.

Short answer: what is Atlan MCP?

Atlan MCP is a Model Context Protocol server that brings metadata capabilities from Atlan into an AI agent's toolbox, aimed at data governance and cataloging. In practice, that means an assistant such as Claude Code or Cursor can work with the descriptive layer of your data estate rather than the raw rows. It sits in the Data Engineering category of the directory and is published from the atlanhq/agent-toolkit repository.

The server follows the usual MCP pattern: the client launches or connects to it, the server advertises what it can do, and the model decides when to call those capabilities during a conversation. If you are new to that model, What Is MCP (Model Context Protocol)? Explained covers the basics of clients, servers and tool calls.

The catalogue entry for Atlan lists compatibility with Claude Code, Claude Desktop, Cursor, Windsurf, VS Code with Copilot, and any MCP client. That breadth matters because a catalog is most useful when it follows you into whichever editor or agent you already use for data work.

What can an agent do with data catalog context?

A data catalog answers questions that sit around the data: what a dataset is, who owns it, how it is described, and how it is governed. Atlan's listing describes its MCP server as exposing metadata capabilities for governance and cataloging, so the value for an agent is context. Before writing a query or a transformation, the agent can look up what the catalog knows instead of guessing from column names.

That changes the kind of request you can make. Instead of pasting definitions into a prompt, you can ask the agent to check the catalog first, then draft code or documentation that matches your organisation's own vocabulary. For analytics engineers, this reduces the back-and-forth of explaining the same tables in every new session.

The exact list of operations is defined by the server itself, so read the repository's README for the current set rather than assuming it. A useful habit is to ask your agent to list the tools the server exposes after you connect it, and compare that list with what your team actually wants an assistant to touch.

How do you connect Atlan MCP?

The listing gives two routes. The recommended one goes through the MCPizy command line: run mcpizy install atlan and the client configuration is handled for you. The alternative is to run the server directly with uvx atlan-mcp-server, which suits people who prefer to manage their own client configuration or want to see exactly what is launched.

1. Install through the CLI with mcpizy install atlan, or start the server with uvx atlan-mcp-server.
2. Follow the repository README for the connection settings your Atlan workspace requires.
3. Restart your client and confirm the server appears in its tool list.
4. Run one harmless lookup before trusting it with anything larger.

If you use VS Code, How to Add an MCP Server to VS Code walks through the editor side of the same process. If the server hangs on startup, the server output is the first place to look, since credentials and connection settings are the usual cause.

What does Atlan MCP not do?

The listing describes metadata capabilities for governance and cataloging. It does not describe a tool for building pipelines, so an agent connected only to Atlan should be expected to read and reason about catalog information, not to orchestrate your data stack. Treat it as one source of context among several.

For transformation work, the dbt server is a separate option: its listing describes managing models, running transformations and querying the semantic layer. The CheckMCP audit of dbt-labs/dbt-mcp gives it 91/100 (grade A), with 616 stars and an Apache-2.0 licence. For ETL and orchestration, Keboola is described as a platform for data workflows, integrations and analytics, and the CheckMCP audit scores keboola/keboola-mcp-server 87/100 (grade B).

Pairing a catalog server with one of these is a reasonable pattern: the catalog supplies meaning, the other server performs the work. For a wider view of the category, Data Warehouse MCP Servers for Analytics Teams compares options by job.

What should you check on access control and data governance?

Metadata is not harmless. Table names, descriptions, owners and classifications can reveal how a business is organised and where sensitive data lives. Whatever credentials you give the server, the agent can see what those credentials can see, so start with an account scoped to the smallest set of assets you need, not an administrator.

Decide in advance which operations you are comfortable letting a model perform without confirmation. Most clients let you require approval for each tool call, and that is a sensible default while you learn how the agent behaves. The guide MCP Server Permissions and Scopes Explained explains how scopes translate into what an agent can actually do.

Also look at the project's health. As of November 2026, the CheckMCP audit of atlanhq/agent-toolkit reports 32 stars, an MIT licence, 70 open issues and a last commit 5 days ago, for a score of 87/100. That is solid with minor gaps in documentation or maintenance, according to the audit. For a method you can repeat on any server, see How to Audit an MCP Server Before Install.

Where can you see the Atlan MCP listing?

The Atlan listing in the directory collects the install commands, the compatible clients, the repository link and the CheckMCP audit in one place. It was last updated on November 4, 2026, and it marks the install as verified. Use it as the starting point, then confirm the details against the repository itself.

The listing shows the audit whether it flatters the server or not, and the full report is available on CheckMCP. If you maintain the server, the listing also offers an optional badge you can paste into your README. It changes nothing about the listing itself.

If you are still deciding between catalog context and hands-on data tooling, compare the Atlan entry with the dbt and Keboola entries side by side. The grades, licences and maintenance signals are published for each, which makes it easier to judge how much weight to put on any one of them before you add it to a daily workflow.