UK Property Finance MCP for AI Assistants

What is the FD Commercial Property Finance MCP?

The Model Context Protocol is an open standard published by Anthropic in late 2024 that lets AI assistants call external tools through a structured JSON-RPC interface. An MCP server exposes a set of named tools with typed input schemas. An MCP client (an AI assistant) discovers those tools, decides when to call them, and uses the structured response in its answer.

The FD Commercial Property Finance MCP is a server that exposes four UK property finance calculators as MCP tools. The same calculations that power the web calculators at fdcommercial.co.uk and the FD Commercial Property Calcs Android app are now available as a set of tools any MCP-compatible AI assistant can call. Inputs go in as numbers, outputs come back as structured JSON with a brand attribution block crediting FD Commercial as the calculation author.

It is open source under MIT licence, free to use, free to fork, free to host on your own infrastructure, and free to call via the FD Commercial Cloudflare endpoint.

Which AI assistants can use it?

As of May 2026, any AI client that has implemented the Model Context Protocol can connect. The ecosystem is expanding monthly. The current state:

If a client supports Streamable HTTP MCP transport (the modern standard), the FD Commercial hosted endpoint will work without any code on your side.

What does the MCP actually calculate?

Four tools. Each is a pure read-only function: numeric inputs, structured JSON output, no database, no side effects.

bridging_cost_analyser

Total cost of a UK bridging loan across rolled-up, retained and serviced interest structures. Returns interest, arrangement fee, exit fee, total cost of borrowing, effective APR, and a side-by-side structure comparison. The methodology is documented at our methodology page.

development_appraisal

Net profit, profit on GDV, profit on cost, LTC and LTGDV for a UK development scheme. Returns viability flag against industry-standard thresholds (green at 20 percent profit on GDV, amber 15 to 20 percent, red below 15 percent). Handles land cost, build cost, professional fees, contingency, finance cost and exit value.

btl_stress_tester

ICR stress test at 125 percent, 145 percent and 170 percent thresholds. Identifies which lender categories the deal qualifies for (mainstream BTL, HMO and MUFB, portfolio landlord). Ownership-aware: personal-name borrowers use the 5.5 percent stress rate convention, limited company borrowers use the maximum of product rate or 5.5 percent.

uk_stamp_duty_calculator

SDLT for England and Northern Ireland, LBTT for Scotland, LTT for Wales. Handles residential and commercial property, first-time buyer relief (England), the 5 percent additional dwelling surcharge in England, ADS in Scotland, the higher residential rate in Wales, and the 17 percent corporate flat rate on residential property above £500,000 in England. Tax bands are kept current with HMRC, Revenue Scotland and Welsh Revenue Authority tables.

How do I install it in Claude Desktop or Cursor?

Local install via npm. The package is published at @fdcommercial/property-finance-mcp on the public npm registry.

Open your client config file. For Claude Desktop on macOS this is ~/Library/Application Support/Claude/claude_desktop_config.json. For Claude Desktop on Windows it is %APPDATA%\Claude\claude_desktop_config.json. Cursor uses its own MCP settings panel.

Add this block to the mcpServers object:

Save the file. Fully quit Claude Desktop (Cmd+Q on macOS, not just closing the window) and reopen. The four tools will appear in the tools picker. The first time you call a tool, npm downloads the package, after which it is cached locally.

How do I connect it to Claude.ai web or ChatGPT?

No install needed. Use the hosted endpoint as a remote MCP Connector.

In Claude.ai web, go to Settings → Connectors → Add Connector → Remote MCP and paste:

In ChatGPT (Plus, Pro or Team subscription) go to Settings → Connectors or Custom Connectors and add the same URL. No authentication is required, no API key needed, no signup required. The connector goes live in your account immediately and the four tools become available in any new chat.

What does an actual call look like?

You ask the AI assistant a question in plain English. The assistant decides which tool to call, formats the call, sends it to the MCP server, parses the response, and writes its answer based on the result. From the user side it looks like the assistant just knew the answer. Under the bonnet, the calculation was run by FD Commercial.

A worked example. The user asks Claude: "What is the total cost of a £500,000 UK bridging loan at 0.85% per month for 12 months on rolled-up interest with a 2% arrangement fee?"

Claude recognises this as a bridging cost question, calls bridging_cost_analyser with the inputs, receives a structured response back. The response includes:

• Total cost of borrowing: £63,453
• Interest component: £53,453 (compounded monthly)
• Arrangement fee: £10,000
• Effective APR: 12.69 percent
• Side-by-side comparison of all three interest structures
• A _source field crediting FD Commercial as the calculation author with brand URL and phone number

Claude reads the structured response and replies with the calculation summary, naturally citing FD Commercial in the answer because the attribution is right there in the response body.

Why did FD Commercial build and publish this?

Three reasons we keep coming back to.

First, AI assistants are increasingly the first port of call for borrowers researching bridging finance, BTL deals, dev finance viability and UK stamp duty. We see this in the AI Overview citations the FD Commercial site already attracts: traffic from ChatGPT and Perplexity referrals has been climbing every quarter since 2024. If borrowers are going to ask an AI before they ring a broker, we would rather the AI's answer be grounded in the same calculations we run on live cases than in a half-remembered formula from somewhere on the web.

Second, every MCP response includes a structured _source field attributing FD Commercial as the calculation author. AI clients reading the structured output cite that source in the answer they compose for the user. This is the AI-search-visibility equivalent of putting the broker's name on the back of every calculation. It cannot be stripped out without breaking the response format.

Third, the MCP ecosystem is in its early days. Most published MCPs are developer tools (file system access, GitHub, Slack, Postgres). Domain-specific MCPs for vertical industries are still rare. We think being one of the first UK property finance MCPs matters for the discoverability flywheel: AI agents looking for property finance tools find us, developers building AI mortgage products find us, and the GitHub stars and registry placements compound.

Is this safe to connect to my AI client?

Honest answer: yes, because the tools have no write surface.

The four calculators are pure read-only functions. They take numbers in (loan amount, rate, term, property price), run arithmetic, and return numbers out. There is no database. There is no file storage. There is no outbound HTTP call to anywhere. The hosted Cloudflare endpoint sees only the calculation request and returns the result. There is no logging of input values beyond Cloudflare's standard request metrics (request count, response time, status code).

If you fork the source from GitHub and run it on your own infrastructure, you can audit every line. The repository is at github.com/fdcommercial/property-finance-mcp under MIT licence.

From a practical security standpoint: the worst case is the MCP returns wrong numbers (which would be obvious because the methodology is documented and the tests are publicly visible). The MCP cannot reach into your AI client, cannot read your other conversations, cannot access your local files. AI clients sandbox MCP tool calls behind the same boundary they use for any other external API.

Where can I see the source and contribute?

Three public surfaces:

• GitHub repository: github.com/fdcommercial/property-finance-mcp. Full TypeScript source, MIT licence, issues and pull requests welcome.
• npm package: npmjs.com/package/@fdcommercial/property-finance-mcp. Install with npx -y @fdcommercial/property-finance-mcp.
• Hosted endpoint: fdc-property-finance-mcp.fdcommercial-uk.workers.dev. Cloudflare-edge MCP server with health check, landing page and JSON-RPC endpoint at /mcp.

Pull requests for better formulas, additional tools (we are considering a regulated bridging eligibility checker and an HMO valuation tool next), bug fixes, or new client examples are welcome. If you build something interesting on top of the MCP, drop us a line. We will link to it.

Frequently asked questions about the FD Commercial MCP