HouseCanary MCP: 149 Property Tools for Any AI Agent

pip install housecanary-mcp

Key Takeaways

• 149 tools across 11 categories — from individual property valuations up to state-level forecasts.
• 36-month forecasts at the property, ZIP, block, block group, metro division, MSA, and state level — exposed natively to any MCP-compatible model.
• Read-only by design. The server filters out write endpoints, so an agent can read the data but can't mutate your portfolios.
• stdio by default, HTTP optional. Works with Claude Desktop out of the box and runs as a hosted endpoint when you need it.

What MCP Is, and Why This Matters

The Model Context Protocol is an open standard for connecting AI models to tools and data. Anthropic published the spec in late 2024, and the major model providers — Anthropic, OpenAI, and Google — have all shipped MCP support since. An MCP server exposes a list of callable tools; an MCP-compatible model (Claude, ChatGPT, an in-house agent) decides when to call them.

For real estate, that changes the workflow. An agent no longer scrapes a listing site or guesses comps. It calls property/value and gets back a valuation with a confidence band and FSD score. It calls zip/market_pulse/timeseries and pulls a historical pulse for the ZIP code.

That's the protocol layer. The data layer matters more — and that's where HouseCanary's AVM accuracy (2.7% postlist, 7.5% prelist median absolute error) becomes the foundation every agent inherits when it grounds itself in our tools.

What's in the HouseCanary MCP

The server wraps the HouseCanary Analytics and Portfolio Monitoring APIs across multiple geographic levels — from individual properties up to states. The full surface, organized by category:

The 36-month forecast horizon you already have through the Data Explorer API is exposed at every geographic level. So is value-by-condition and the full Property Explorer link API.

Install and Run

The package is on PyPI:

pip install housecanary-mcp

Or run it without installing, using uvx:

uvx housecanary-mcp

Authenticate with the same credentials you use for the HouseCanary API:

export HOUSECANARY_API_USERNAME="your-username"
export HOUSECANARY_API_PASSWORD="your-password"
housecanary-mcp

To wire it into Claude Desktop, add this to claude_desktop_config.json:

{
  "mcpServers": {
    "housecanary": {
      "command": "uvx",
      "args": ["housecanary-mcp"],
      "env": {
        "HOUSECANARY_API_USERNAME": "your-username",
        "HOUSECANARY_API_PASSWORD": "your-password"
      }
    }
  }
}

To run it as a hosted HTTP endpoint instead of stdio:

FASTMCP_TRANSPORT=http \
FASTMCP_HOST=127.0.0.1 \
FASTMCP_PORT=8000 \
FASTMCP_STREAMABLE_HTTP_PATH=/mcp \
uvx housecanary-mcp

That's the full integration. The default endpoint is http://127.0.0.1:8000/mcp, ready for any MCP-compatible client.

Note on billing. Tool calls count as normal HouseCanary API calls and are billed against your existing plan or contract — there's no separate MCP fee. Portfolio Monitor, Market Pulse, and the Bulk Property Data endpoint are no-additional-charge (subscription products, included in your plan). Everything else (Property, ZIP, Block, MSA, State, Property Explorer) is billed at your contract rate. AI agents can call tools many times in a single workflow, so monitor usage in your HouseCanary billing dashboard during the first week to set an expectation.

How It Compares

Two other property data vendors have shipped MCP servers this year: ATTOM and Cotality (formerly CoreLogic). The three approaches are different.

Each vendor optimized for a different buyer. ATTOM bundled MCP with Databricks for bulk-data analytics teams. Cotality leans on enterprise governance and the open-source OSI semantic standard. HouseCanary's MCP wraps the deepest analytics surface in the category and ships as a normal Python package — install it, point it at credentials, and any MCP client can call it.

The forecast point matters. Predictive analytics is HouseCanary's structural data advantage — 36-month value forecasts, propensity-to-list scoring, ZIP-level rental forecasts. None of that shows up in a competitor's MCP today. With this release, an AI agent can ground itself in HouseCanary's forward-looking models, not just point-in-time data.

What This Enables

A few concrete agentic workflows this server makes possible today:

• Underwriting assistants. A loan officer's copilot pulls property/estimate, property/value_fsd_threshold, property/ltv_origination, and property/flood in a single conversation, then summarizes risk for the file.
• Investor deal screens. An investor's agent runs property/value, zip/hcri for gross rental yield, block/hazard_* for disaster exposure, and msa/hpi_ts_forecast for the 36-month outlook before recommending a bid.
• Portfolio monitoring at conversational level. A servicer asks "which assets in portfolio 12345 are now above 80% LTV?" and the agent calls portfolio-monitor/portfolios/{portfolio_id} plus property/ltv_details to answer with current numbers.
• Listing-presentation prep. An agent's CMA tooling pulls property/value_analysis, property/sales_history, and zip/market_pulse/latest to build a defensible price discussion for a kitchen-table conversation.

These aren't future use cases. They work today, against the live HouseCanary API, through any MCP-compatible model.

Why Ship This Now

Property data has been hard for AI to use. APIs are fragmented, schemas are inconsistent, and models that try to reason about real estate without grounding produce confident but wrong answers — bad comps, hallucinated owners, AVMs invented from thin air.

MCP is the standard that fixes that. It gives any model a clean way to call data that's accurate, documented, and auditable. By publishing the HouseCanary MCP as an open Python package on day one, we're making the lowest-friction path to grounded AI in real estate a HouseCanary path.

"Every AI agent in property tech needs accurate data underneath it. Shipping our MCP server as a public, installable package — wrapping 149 endpoints and 36-month forecasts — means any builder can ground their agents in HouseCanary's data the same day they think to ask. That's how the next generation of real estate AI gets built."

— Chris Rediger, CEO

FAQ

Do I need a new API key?

No. The MCP server uses your existing HouseCanary API username and password. If you already have access to the Analytics API, you have access to the MCP server.

How is MCP usage billed?

Every tool call counts as a normal HouseCanary API call and is billed against your existing plan or contract — there's no separate MCP fee, but agents can call tools many times in a single workflow, so usage adds up. Three exceptions: Portfolio Monitor, Market Pulse, and the Bulk Property Data endpoint don't carry per-call charges (they're subscription products, included in your plan). All other endpoints (Property, ZIP, Block, MSA, State, Property Explorer) are billed at your contract rate. Track current usage in your HouseCanary billing dashboard — that's the quickest way to see usage for the current billing period.

Is the source code open?

The package is published as an installable wheel and source distribution on PyPI. The license is proprietary — you can install and run it, but the underlying API access is governed by your HouseCanary contract.

Which AI models can use this?

Anything that speaks MCP. That includes Claude (Desktop, Code, API), Cursor, Cline, Continue, ChatGPT desktop with MCP support, and any custom agent built on the official Anthropic, OpenAI, or Google MCP client libraries.

Is it stdio or HTTP?

Both. Default transport is stdio for local clients like Claude Desktop. Set FASTMCP_TRANSPORT=http to run it as a hosted HTTP endpoint on port 8000.

What's excluded from the MCP surface?

POST methods and the component_mget batch endpoints are filtered out by design. The MCP server is read-only — agents can query but not mutate state.

How does it compare to building directly against the REST API?

The REST API is still the right choice for high-throughput backend workloads. MCP is for agentic and conversational use cases — wherever an LLM is doing the calling.

Get Started

The package is live on PyPI: pypi.org/project/housecanary-mcp.

pip install housecanary-mcp

If you don't have HouseCanary API credentials yet, start with the Analytics API — every endpoint there is now also an MCP tool.

Already a self-serve customer? Check pricing for your current plan and your current bill in your billing dashboard.