Dilix MCP
Open-source MCP server providing real estate regulatory intelligence (zoning, permits, entitlements, deal scoring) for US properties, enabling AI agents to access 10 callable tools.
README
Dilix MCP
The open-source MCP server for Dilix — the regulatory intelligence layer for real estate AI agents.
Add Dilix to any MCP-compatible AI agent in 30 seconds. Get zoning, permits, entitlement roadmaps, state legislation flags, and deal scoring for any US property — surfaced as 10 callable tools.
Heads up: This package is the open-source client. The Dilix API (data + intelligence) requires an API key — get one free at dilix.ai/api-keys.
Why Dilix exists
Real estate is one of the largest opaque asset classes in the world. Owners close on properties unaware of regulatory shifts that compress NOI. Flippers burn months of carry on permits that never land. Brokers waste deal cycles chasing entitlements that won't clear.
Dilix is the data layer that catches it all — every permit, zoning change, and bill across 16 cities and 11 states, going national. For humans, and the AI agents reshaping the industry.
This MCP package gives those agents direct access.
Install
npm install -g @dilix/mcp
Or run on demand without installing:
npx @dilix/mcp
Get an API key
Free tier: 100 calls/month. Sign up at dilix.ai/api-keys.
Set it as an environment variable:
export DILIX_API_KEY="your-key-here"
Wire it into Claude Desktop
Add to ~/.claude/claude_desktop_config.json (Mac) or
%APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"dilix": {
"command": "npx",
"args": ["-y", "@dilix/mcp"],
"env": {
"DILIX_API_KEY": "your-key-here"
}
}
}
}
Restart Claude Desktop. The Dilix tools will appear in the MCP tool palette.
Wire it into Cursor / Continue / custom agents
Any MCP-compatible host. Spawn the binary with DILIX_API_KEY in the
environment. Speaks standard MCP over stdio.
Available tools
| Tool | What it does |
|---|---|
analyze_zoning |
Zoning + HCD compliance + state-law flags for any address |
get_entitlement_roadmap |
Full approval pathway with timelines, fees, risks |
lookup_property |
Property records, valuation, sale history (US) |
find_opportunities |
Scan a city for SB-9, SB-79, AB-2011, AB-130 candidates |
scan_portfolio |
Bulk compliance + opportunity scan across up to 50 properties |
watch_parcel |
Hourly autonomous monitoring with email alerts |
track_permit |
Real-time permit status + status-change notifications |
analyze_deal |
Full deal score: zoning + property + financials + verdict |
source_deals |
Source candidate properties matching investment criteria |
get_market_rates |
Live SOFR + 10Y Treasury for cap-rate underwriting |
Full schemas: docs/tools.md (coming soon — for now, MCP hosts will expose them automatically once connected).
Example: Claude analyzing a deal
User: Should I pursue 555 California St for a residential conversion?
Claude (with Dilix MCP):
1. analyze_zoning(city="San Francisco", address="555 California St")
→ C-3-O zoning · Builder's Remedy active · AB-2011 eligible
2. get_entitlement_roadmap(...)
→ Ministerial pathway · CEQA exempt · 6-9 mo timeline · $X est. fees
3. lookup_property(address1="555 California St", address2="SF, CA 94104")
→ Built 1969 · 350K sqft · last sold 2018 · $X assessed
Verdict: Strong candidate. Pursue. Submit SB-330 vesting application
within 30 days to lock current rules.
Grounding — every tool returns reasoning + sources
Starting in v1.1.0, every Dilix MCP tool surfaces a structured envelope to the host model:
_reasoning— plain-English narrative the agent can quote verbatim to the user. Composed by Dilix from structured fields, not extrapolated. Eliminates the most common failure mode (host model invents a cap rate / code section / date that wasn't in the response)._sources— citations the agent can surface as "Source: X" links. Includes provider, citation text, and a verifiable URL where available. Pulled from real authorities: ATTOM, DataSF, FEMA, HUD, NY Fed, FRED, Cal. Civ. Code, municipal codes, etc._caveats— soft warnings the agent should pass through (e.g. "no parcel-level zoning available for this address — answers are city-level only")._schemaVersion— pin behavior across upgrades.
The MCP server splits these into separate content blocks so the host model receives the narrative as the primary text and citations as a structured second block. Agents that want the raw payload still get it as a JSON block.
Endpoints currently emitting the envelope: analyze_deal, analyze_zoning,
get_entitlement_roadmap, lookup_property, market_rates,
find_opportunities, scan_portfolio. Others fall back to the legacy
single-block JSON response.
Configuration
| Env var | Required? | Default | Description |
|---|---|---|---|
DILIX_API_KEY |
Yes | — | Get from dilix.ai/api-keys |
DILIX_API_BASE |
No | https://api.dilix.ai |
Override for self-hosted or staging |
Pricing
The MCP client is free and open-source (MIT). The hosted Dilix API is usage-based:
- Free tier: 100 calls/month
- Pro: $49/month — 5,000 calls
- Agentic: $499/month — 50,000 calls + portfolio monitoring + webhooks
- Enterprise: custom — email ops@dilix.ai
See dilix.ai/pricing for the live tiers.
Why open source?
Same playbook as Anthropic, Vercel, Supabase, Resend: open the developer surface, charge for the data + infrastructure behind it.
We open-sourced the MCP client because:
- Devs deserve to read the code that runs in their agent
- The MCP spec is open (Anthropic, late 2024) — closing the client would be hypocritical
- The client is just the "USB cable." Our actual value is the data + scoring + intelligence behind the API
Project status
Maintained best-effort by a solo founder. See STATUS.md for expectations and how to help.
For commercial support, custom integrations, or enterprise SLAs: ops@dilix.ai
Security
Found a vulnerability? See SECURITY.md — please email
ops@dilix.ai (subject: [SECURITY]) rather than opening a public issue.
Contributing
PRs welcome. Small and focused win. Large refactors usually don't.
git clone https://github.com/dilix-ai/mcp.git
cd mcp
npm install
npm run build
npm run dev
License
MIT. See LICENSE.
"Dilix" is a trademark of Ownership Theory LLC. The MIT license grants rights to the source code, not to the brand.
Links
- 🌐 Website: dilix.ai
- 🔑 Get an API key: dilix.ai/api-keys
- 📖 Docs (full): dilix.ai/agents
- 📰 Weekly Briefing: dilix.ai/briefings
- 📧 Contact: ops@dilix.ai
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。