As we wrote in our piece on using AI for planning analysis, AI results can mislead without reliable data underneath. The CommunityScale MCP server lets agentic AI tools like Claude and ChatGPT query the same housing data that powers our dashboards. This guide will walk you through the onboarding process.

What the Model Context Protocol is

The Model Context Protocol (MCP) is an open standard that lets large language models connect securely to external data sources and tools. Instead of relying on whatever a model absorbed during training, an MCP-connected assistant calls live tools, gets governed query results back, and can show where each number came from. MCP is specifically suited for self-serve data analysis because not only does it deliver current and actionable data, it also provides a semantic layer that lets your AI know how to use the data responsibly.

What our server offers

Connecting gives your assistant with these capacities:

• Our computed forecast figures for municipalities and counties: housing supply targets, cost burden by income group, affordability gaps, AMI thresholds, and population projections, matching what the CommunityScale dashboards display.
• Source data, such as survey-weighted ACS microdata. Cross-tabulate households or people by tenure, race, age, income band, structure type, commute mode, and dozens of other dimensions, with every estimate properly weighted.
• Published CommunityScale reports. Pull the full narrative report for any covered place as clean markdown.
• Our methodology. The server ships a metric glossary, standard analysis recipes, and a list of data gotchas, so the assistant analyzes housing data the way we do.

Every request resolves to a governed metric definition, and every response includes provenance identifying exactly which sources were used to produce the numbers.

Five tools

• findEntity resolves a place name or FIPS code to a CommunityScale geography. It runs first and feeds the others.
• getForecast returns our computed dashboard figures for a place, by topic: summary, housing supply, cost burden, affordability, AMI, or population.
• exploreData runs open-ended analysis over our exhaustive primary data sources, more information on that soon.
• dataSchema lists the dimensions and measures exploreData accepts, so the assistant can discover what is queryable.
• getReport fetches the published dashboard report for a municipality or county as markdown.

Connect in about a minute

The endpoint is https://api.communityscale.io/mcp/v1 and authentication is OAuth. Your AI tool opens a browser window, you create or sign in with your CommunityScale account, and you’re connected.

Claude on the web and desktop

1. Open Claude’s connector settings (button below).
2. Click Add custom connector.
3. Name it CommunityScale and paste https://api.communityscale.io/mcp/v1 as the URL.
4. Click Connect and sign in when prompted.

Claude Code

One command in your terminal:

claude mcp add --transport http communityscale https://api.communityscale.io/mcp/v1

Then type /mcp inside Claude Code to sign in. The tools are available in every session afterward.

ChatGPT

1. Open ChatGPT’s settings and go to Connectors (button below).
2. Under Advanced, enable Developer mode.
3. Click Create, name the connector CommunityScale, and paste https://api.communityscale.io/mcp/v1.
4. Sign in through the OAuth prompt.

Custom connectors require ChatGPT’s developer mode. Our tools are all read-only, the category ChatGPT supports on Plus and Pro plans as well as Business, Enterprise, and Edu workspaces.

Codex CLI

Two commands: one to add the server, one to sign in.

codex mcp add communityscale --url https://api.communityscale.io/mcp/v1
codex mcp login communityscale

Your own agents

Anything that speaks MCP can connect. Both the Claude API and the OpenAI Responses API accept remote MCP servers by URL, so a custom agent gets the same five tools by pointing at the endpoint and completing the OAuth flow.

What to ask once you’re connected

• “What share of renter households in Worcester, Massachusetts spend more than half their income on housing?”
• “How many homes does Rockland, Maine need to add by its target year, and what drives that number?”
• “Compare homeownership rates by householder race across the New England states.”
• “Pull the CommunityScale report for Chittenden County, Vermont and summarize the three biggest issues.”

In each case the assistant resolves the place, picks the right tool, and quotes numbers that match our dashboards exactly, along with the methodology notes to interpret them correctly.

The same data behind our consulting work

We focus on housing supply because it is what communities can act on: zone for it, plan for it, build it. The MCP server takes the evidence base we bring to housing plans and zoning work and makes it available wherever you already work, whether that’s a chat window, a code editor, or an agent your team built. Reach out if you would like access for your team or your community.