Esri MCP Server: Connect GIS Workflows to an AI Assistant

An MCP server for Esri connects an AI assistant to ArcGIS Online, ArcGIS Enterprise, or ArcGIS Pro so the assistant can query geographic data, read layer metadata, and reason about analysis results without a manual copy-and-paste step. For GIS professionals, the practical gain is an assistant that understands the data you are looking at rather than an abstract description of it. Tools like fenbs keep the map sources, project constraints, and accepted analysis decisions separate from the GIS system itself, where any connected AI can find them.

Esri is the company behind ArcGIS, the dominant GIS platform used in government, engineering, urban planning, and environmental science. If your work involves spatial data, an MCP connection reduces the time spent moving information between a GIS platform and an AI assistant.

What an Esri MCP server can expose

ArcGIS provides REST APIs covering feature layers, map services, geoprocessing tasks, and portal items. An MCP server wraps those endpoints as named tools. Common capabilities include querying a feature layer for attributes and geometry, listing layers in a web map, reading metadata about a hosted dataset, and running a published geoprocessing service.

The access level depends on the ArcGIS account or API key used. Public layers need no credentials. Private organizational layers need a token scoped to the layers involved. Check the server documentation to confirm which ArcGIS product it targets and whether it supports both ArcGIS Online and Enterprise deployments.

Evaluating an Esri MCP server

Search the MCP server directory for ArcGIS or Esri MCP tools. When evaluating one, check whether it supports the ArcGIS version your organization uses. ArcGIS Online and ArcGIS Enterprise use similar REST endpoints but may differ in authentication methods and available services.

Also check whether the server exposes read tools only or includes write tools that can update or delete features. For most analysis workflows, read-only access is the safe starting point. Write access to a shared feature layer affects every user and every connected system that reads that layer.

Setting permissions for a GIS connection

Create a dedicated ArcGIS account or API key for the MCP connection with the minimum privilege needed: typically view access to the layers and services involved in the project. Do not use an account with edit rights to shared layers unless the workflow explicitly requires AI-assisted editing and a person has approved each change.

Record the permission decisions before you connect. Which layers may the assistant read? May it run geoprocessing services, or only query data? Does it need access to the organization's full portal, or only a specific web map? These answers belong in a shared record that the whole team and every connected AI can see.

A Decisions and rules entry captures those boundaries once, so a new team member or a new AI session starts from the same approved scope rather than discovering the limits through trial and error.

Keeping map sources and analysis assumptions organized

GIS projects often depend on a specific set of authoritative data sources: the official parcel layer, the approved floodplain dataset, the current road centerline file. Using a different version of one of those layers produces different results. Recording which sources the project uses and where they come from is as important as recording what the analysis found.

Analysis decisions matter too. If the project uses a specific buffer distance because of a regulatory requirement, or excludes a particular area because of a known data gap, those choices need a home outside the GIS platform so they reach the next analyst and the next AI session.

Typical analysis workflows with an AI assistant

Ask the assistant to query a feature layer for all parcels within a specific zoning code and summarize the attribute distribution. Or ask it to compare two layers for overlap and explain where they diverge. Keep each query narrow: one layer, one question, one time range. Wide queries against large datasets can time out or produce more data than the assistant can reason about usefully.

Treat the assistant's output as a starting point for human review, not a finished analysis. Spatial reasoning requires checking that queries returned the expected features and that summary statistics reflect the actual distribution. The assistant helps you navigate the data faster; the analyst confirms the result is valid. For general MCP setup guidance, see mcp-server-configuration.

How to do this with fenbs

1. Create a free account and set up a project for the GIS work or the specific analysis effort.
2. Add Links and listings for the authoritative data sources: the layer name, the portal URL, the version or date, and the source organization.
3. Record the analysis boundaries as Decisions and rules: which layers are in scope, which areas or periods are excluded, and which processing choices were approved.
4. Connect your AI assistant over MCP from Integrations. Ask it to read the project context before querying any GIS data, so it knows the approved sources and scope.
5. After each analysis session, record the key findings as context notes and any new scope decisions as Decisions and rules so the next session starts from the current state of the analysis.