Enterprise
Esri Ships Native MCP Server in ArcGIS Enterprise 12.2, Letting Claude and Copilot Studio Query Live GIS Data
ArcGIS Enterprise 12.2 ships 11 built-in MCP tools that give compatible AI clients, including Claude Desktop and Microsoft Copilot Studio, direct read access to an organization's spatial data, mapping layers, and 3D scene content.
Mainstay Communications Team · · 5 min read

In brief
Esri has shipped Model Context Protocol support in ArcGIS Enterprise 12.2, turning a production GIS deployment into a live data source for foundation models. Compatible AI applications, including Claude Desktop, Copilot Studio, GitHub Copilot, ChatGPT, and Gemini CLI, can now discover and invoke an organization's geospatial tools through 11 built-in tools covering portal search, mapping, location services, and geoprocessing. Access runs entirely through ArcGIS's existing security model: agents see only what the authenticated user is permitted to see.
The Gap Esri Is Closing
Esri, which dominates the enterprise GIS market with its ArcGIS platform, has watched foundation models grow in capability while remaining spatially illiterate. Ismael Chivite, Esri's distinguished product manager for geospatial AI, put it plainly: "There are many agentic platforms out there that do amazing things, but they are not geospatially intelligent." He described the problem as "a big business risk" for organizations building agentic AI on tools that ignore geography.
The ArcGIS Enterprise MCP server is the company's structural answer. Esri implemented the open Model Context Protocol standard and published a versioned endpoint that any compliant client can discover and query, so one implementation serves every AI platform. The server implements MCP specification version 2025-06-18 and runs as a separate web application installed on the ArcGIS Enterprise hosting server.
Eleven Tools, Four Categories
The default catalog ships with 11 built-in tools in four categories: content and data, mapping, location services, and geoprocessing. Named built-ins confirmed in Esri's documentation include Search Portal Content, Describe Item, Describe Layer, and Get GP Task Job Status, the last used for tracking asynchronous geoprocessing runs.
Administrators can extend the catalog by exposing custom geoprocessing tools without writing MCP-specific code. Custom catalogs, introduced in this release, let admins curate a scoped subset of tools and share it only with specific users and groups, limiting what a given AI client can reach.
A parallel Esri MCP server for the hosted ArcGIS Location Services platform, in beta since June 2026, ships a separate set of 9 tools: forward and reverse geocoding, route and service-area solving, elevation queries, static map rendering, demographic data enrichment, and geometry buffering.
Which Clients Work Today
Esri's FAQ names 7 AI clients as supported examples for the Enterprise server: Microsoft Copilot Studio, Anthropic's Claude Desktop, Kiro, Google's Gemini CLI, ChatGPT, GitHub Copilot, and Goose. Clients must support remote MCP connections over Streamable HTTP and one of Esri's supported authentication methods. Tools that connect only via the older stdio transport require an additional bridge to reach the ArcGIS server.
Spatial Data and 3D Scene Content
Because the MCP server can search and describe portal items, an AI agent can locate web scenes, 3D feature layers, Gaussian splat scene layers, and imagery layers stored in the organization, retrieve their metadata, and pass that spatial context into a model's reasoning chain.
ArcGIS Enterprise 12.2 extended the platform's 3D capabilities alongside the MCP launch. The release added Google's Photorealistic 3D basemap, in preview, as a new option in the Scene Viewer basemap gallery. Gaussian splat layers, which encode scene geometry as point clouds of colored splats, now support elevation profiling and line-of-sight analysis in this version. That 3D data lives in the same portal the MCP server indexes, making it discoverable by connected agents.
Permissions and Governance
The design keeps an organization's existing permission structure intact. External AI client access is disabled by default in Enterprise 12.2; an administrator must enable it before any agent can connect. The agent then operates within the authenticated user's ArcGIS sharing permissions, with no separate access control layer to maintain.
The initial release is read-only. Write operations, including feature editing and service publication, aren't supported in this version. Chivite has described the deliberate scope limits as intentional, noting that Esri's internal AI Governance Board reviews AI features before release.
Esri says the MCP connection itself adds no additional charge. Normal ArcGIS service billing applies to underlying resources the agent invokes, including ArcGIS Online credits for some location service calls.
Generally Available, Independent Release Cycle
Esri lists MCP for ArcGIS Enterprise as generally available in 12.2, which sets it apart from the release's AI assistants and AI agents, features that ship as beta and get support through the ArcGIS Enterprise Early Access Community. On Windows and Linux, MCP arrives as a separate installable component, MCP Services for ArcGIS Enterprise, with its own versioning and release lifecycle, so its capabilities can update between ArcGIS Enterprise upgrades. On Kubernetes, it's bundled with ArcGIS Enterprise 12.2 and later.
Blue Raster, an Esri partner that hosted a webinar preview of the integration, ran a demo showing an MCP-connected agent narrowing 43,032 land parcels in Delaware County, New York, to 2 qualifying sites, displaying its reasoning at each filtering step. For GIS teams whose data infrastructure is already on ArcGIS, 12.2 means foundation models can reach that investment without a custom integration project.
The Enterprise MCP server is generally available in 12.2. The hosted MCP for ArcGIS Location Services, a separate product on ArcGIS Location Platform, remains in beta, with no general availability date announced.
Sources
- MCP for ArcGIS Enterprise: Geo-Enabling External AI Applications, Esri Blog
- What's New in ArcGIS Enterprise 12.2, Esri Blog
- AI Features Now Available in ArcGIS Enterprise 12.2, Esri Blog
- What's New in Geospatial AI, November 2026, Esri Blog
- MCP in ArcGIS Enterprise Introduction, Esri Documentation
- MCP in ArcGIS Enterprise FAQ, Esri Documentation
- Administering MCP in ArcGIS Enterprise, Esri Documentation
- ArcGIS Location Services MCP Tools, Esri Developers
- MCP Support (beta) in ArcGIS Location Platform, Esri Blog
- Recapping Blue Raster's MCP for ArcGIS Webinar, Blue Raster
- ArcGIS MCP: Let AI Assistants Read Your GIS Safely, GeoLink
- Agentic AI Has to Know Where It Is, xyHt
Companies mentioned
Mainstay Digital · Spatial and 3D visualization for industrial companies

