Get Started

This page walks you through the important topics for getting up and running with the Model Context Protocol (MCP) catalog in ArcGIS Enterprise. It is recommended that you follow this page in order as you get started with testing—deploy the overlay, choose how you will authenticate, connect a client, prepare your data, optionally expose your own tools, watch the demo to see it all in action, troubleshoot using ArcGIS Server logs, and remove the overlay.

If you are new to MCP and want to understand how it fits into the context of ArcGIS Enterprise, see Introduction to MCP in ArcGIS Enterprise.

Each section below summarizes a step and links to its full topic.

Apply the MCP Overlay

The MCP overlay adds the platform catalogs web application to your ArcGIS Server, standing up the /platform context that serves the MCP catalog. This is the first step: the MCP catalog endpoint does not exist until the overlay is deployed. You can deploy it with the provided installation script or manually, then confirm the platform catalogs page loads.

See Apply the MCP Overlay.

Supported Authentication Techniques

The MCP endpoint is protected, so you need a credential before you can reach it. Three techniques are supported—API key, token, and OAuth 2.0 (manual). Pick the one that fits how you plan to connect and obtain the credential.

See Supported Authentication Techniques.

MCP catalog endpoint

To connect an MCP-compatible AI client to the MCP catalog enabled in ArcGIS Enterprise (after the MCP overlay is applied successfully), use the following endpoint URL:

https://<domain name>/<Web Adaptor name or context name>/platform/mcp

For example:

https://mcpbeta.webgistesting.net/server/platform/mcp

Connect from the MCP Client

Register the MCP catalog endpoint in any MCP-compatible client—such as Microsoft Copilot or Claude Desktop—and supply your credential. The topic includes short video walkthroughs for Copilot (API key/token and OAuth 2.0) and Claude Desktop, and the same pattern applies to other clients like ChatGPT, Cursor, GitHub Copilot, and Postman.

See Connect from MCP Clients.

Prepare Your Data to Be MCP-Ready

MCP tools only work with content you explicitly opt in. Tag the portal items you want to expose with mcp, and give them rich titles and descriptions so the tools can discover the right data for a given prompt.

See Prepare Your Data to Be MCP-Ready.

Expose Custom GP Tasks as Custom MCP Tools

Beyond the default tools, you can surface your own geoprocessing tasks—synchronous or asynchronous—as custom MCP tools. Make the Geoprocessing service MCP-ready and tag it mcp, and it appears as a new tool in the default catalog.

See Expose Custom GP Tasks as Custom MCP Tools.

MCP Tools Catalog

With your data prepared, the MCP tools can discover and act on it. The catalog lists every available MCP tool—across content and data, mapping, location services, and geoprocessing—each with its name, description, schema, and sample prompts.

See MCP Tool Catalog.

Watch the demo

Watch the short demo video to understand how MCP tools work in an ArcGIS Enterprise environment—how an AI client discovers the available tools, how the tools are invoked behind a natural-language prompt, and what kind of results to expect.

If the embedded player does not load in your browser, you can also download the video.

Troubleshoot MCP Using ArcGIS Server Logs

When something isn't working—a client can't connect, a tool call fails, or results look wrong—the MCP catalog writes diagnostic entries into the ArcGIS Server logs. You can query them through the ArcGIS Server Administrator Directory or the Admin API to find the cause.

See Troubleshoot MCP Using ArcGIS Server Logs.

Remove the MCP Overlay

If you need to roll back—for example, after testing—you can undeploy the overlay. This removes the arcgis#platform web application and its context descriptor and cycles the ArcGIS Server service, so the /platform context is no longer served. You can remove it with the provided scripts or manually.

See Remove the MCP Overlay.

Once you have connected a client and run a few prompts, post any feedback, questions, or issues in the Early Adopter Community (EAC) forum so the team can follow up with you.