Skip to content

02.04 / Grow

Agent surfaces.

An MCP server, a public JSON API, an OpenAPI spec and a markdown twin of every page, all generated from the same data the site renders, plus a listing in the registries where assistants look for tools. The pages people read do not change. Software gets a second way in.

Book a callTell us about your project

In detail

What you should know.

An older man reads his phone in a dim kitchen, a moka pot heating on the gas flame beside him and a narrow city street outside the window

For example: 22 public routes prerendered as static HTML. Hotel Marien

Book a callTell us about your project

What it is

Four ways to consult your website as software, plus the paperwork that makes them findable: an MCP server, a public JSON API, an OpenAPI specification, a markdown twin of every page, and a listing in the registries where assistants and their users go looking for tools.

Four ways an assistant can reach the same answer

Four ways an assistant can reach the same answerAn assistant can call a Model Context Protocol tool, fetch the JSON API, read the markdown twin of a page, or crawl the HTML. All four are generated from the same content, so none of them can disagree with the others.An assistant, answering a buyer"who builds hotel sites?"MCPasks a tool a questionJSONreads services and cases asdata.mdreads any page as plain textHTMLcrawls the page like a searchengineOne source of contentso the four never disagree
A site you can only crawl competes against a site you can question.

The pages do not change. Everything a person sees stays exactly where it is. What gets added is a second way in, built for software.

Why a crawled site loses to a queried one

When someone asks ChatGPT or Claude for a supplier, the assistant answers with what it can consult. Usually that means whatever HTML a crawler happened to fetch, some weeks ago, without running your JavaScript.

A site with agent surfaces is in a different position. The assistant calls a tool, asks for your prices, gets them back as data in the language it asked in, and quotes them. That is the difference between being remembered vaguely and being read exactly.

What gets built

The MCP server

MCP is the protocol assistants use to call tools. Your server exposes a small read-only set: list what you sell, get the price list, find a page, return one page in full. Read-only means it can face the open internet without becoming a way into anything else you run.

The JSON API and the OpenAPI spec

The same content over plain HTTP, no key. The OpenAPI file describes every endpoint, so a developer, or an agent writing its own client, can use it without asking you a single question first.

A markdown twin of every page

Each page also answers at the same path with a .md suffix: front matter, the heading, the text, no navigation and no scripts. It is what an assistant actually wants when it fetches a page, instead of a template it has to dig through.

Getting listed

The server and the API are submitted to the public MCP registries and tool directories. A tool nobody can find is a tool nobody calls.

How it stays true to the site

The surfaces are generated from the same data the pages are built from. That is the whole trick: they cannot drift out of sync, because there is no second copy of the content to keep updated. And no tool takes a URL from the caller, so the server never turns into an open proxy for somebody else's traffic.

The proof is our own site

polargate.ai runs these surfaces in public. Checked on 3 September 2026: the MCP server at /api/mcp answers a tools/list call with seven read-only tools, /api/v1 returns the site as JSON with no key and no login, /openapi.json describes that API, and every page answers at the same path with a .md suffix, across every URL in the sitemap that day.

Point your assistant at it and read what comes back. That is the same thing your buyers will do to you.

What this is not

It is not a replacement for search. The crawlers behind answer engines still fetch raw HTML and still do not run JavaScript, so the pages have to exist as real HTML first. It is not a guarantee of citations either: no one can sell that. What it does guarantee is that when an assistant tries to find out what you sell and what it costs, there is a precise answer waiting instead of a guess.

What you get

What you take away.

Everything is written down before we start, and everything stays in accounts your company owns.

01Deliverables

  1. 01

    A public MCP server with a small set of read-only tools over your own content

  2. 02

    A JSON API with a stable URL per resource, no key and no login

  3. 03

    An OpenAPI specification, so a developer or an agent can build its own client

  4. 04

    A markdown twin of every page at the same path, with front matter and no scripts

  5. 05

    Submission to the public MCP registries and tool directories

  6. 06

    Everything generated from the same data as the pages, so the surfaces cannot fall out of sync

  7. 07

    Read-only by design: no tool accepts a URL from the caller, so the server cannot be used as a proxy

02How it runs

  1. 01 · Week 1 to 2

    Audit and baseline

    We fetch your site the way GPTBot, Claude, PerplexityBot and Googlebot fetch it, read Search Console and analytics, check the structured data against the visible text, and run a panel of buyer prompts to see where you appear today. The output is a written baseline and a ranked list of findings.

  2. 02 · Week 3 to 6

    Fix at the source

    We work in the repository: real HTML per route, the schema.org graph, hreflang, sitemap, robots and page speed. Every fix ships as a pull request with the test that proves it, so the same problem does not come back on the next deploy.

  3. 03 · Month 2 onwards

    Content and campaigns

    We publish focused pages that each answer one buyer question, in every language you sell in, and run paid advertising where it is the fastest way to test a message. Volume is set by the retainer, so you know in advance what a month contains.

  4. 04 · Every month

    Measure, report, repeat

    Each month we re-run the audit and the prompt panel, compare against the baseline, and report clicks, impressions, average position and assistant referrals in your client portal. Then we agree the next month's list with you before touching anything.

For example: 16 indexable URLs, Spanish and English. Hotel Talavera

Book a callTell us about your project

Who it is for

We don't take every project.

We would rather tell you now than halfway through. If you recognise yourself on the left, let's talk.

It fits if

  • Your site already exists and your buyers search on Google or ask ChatGPT
  • You want the fixes made in the code, not a keyword report
  • You are happy to measure every month against a recorded baseline
  • You can give access to the site, Search Console and analytics

It is not for you if

  • You want a guaranteed first position or a guaranteed mention in ChatGPT
  • You want to buy links or publish unreviewed text in bulk
  • You expect results within a week
  • Nobody can touch the code of your site, and there are no plans to change that

For example: 105,212 characters of body text readable by AI crawlers across 65 URLs, up from zero. Malacca Tables

Book a callTell us about your project

Timing and budget

How we quote.

No price list and no hourly billing during a build. After a short call you get a written quote for your project.

Timeline
When Polargate builds the site, the surfaces ship with it and are not a separate phase. On a site that already exists, the work starts with a short review of where the content actually lives. If the pages are already generated from structured data, the surfaces are mostly wiring. If the content only exists inside the templates, that structure has to be built first, and the review says which of the two situations you are in before anything is quoted.
Team
Our team builds the surfaces; Andrés Ciordia, Chief AI Officer, leads the AI side and Pedro Ciordia, CTO, reviews them before they go live. They are tested the way a buyer would test them: by pointing a real assistant at the live server and reading what it answers.
How the price is set
Agent surfaces are included when Polargate builds the site. On a site that already exists, a fixed-price add-on quoted after the review of what the site publishes today.
The stack
ViteReact

For example: 98 KB mobile hero image, down from about 288 KB. Hotel Marien

Book a callTell us about your projectAsk for your quote

Work

Work in this line.

Real projects. Open each one to see what we built and the figures we measured.

For example: 27 published FAQs with FAQPage schema. Hotel Talavera

Book a callTell us about your projectAll work

Same line

Other services in Grow.

Each service has its own page with what it includes, the usual timeline and the projects where we did it.

For example: 65 routes pregenerated at build time with their own metadata and JSON-LD. Malacca Tables

Book a callTell us about your project

FAQ

What people ask us.

What is an MCP server, and why would my website need one?
MCP is the protocol assistants like Claude use to call tools. An MCP server on your site turns your content into a handful of read-only tools an assistant can call: list what you sell, get the prices, fetch one page in full. Crawling gives the assistant whatever HTML it happened to keep. A tool call gives it your own data, in your own wording, at the moment it asks. Polargate's server is public at polargate.ai/api/mcp, with seven read-only tools as of 3 September 2026, so you can point your assistant at it and judge the idea on the result rather than on the description.
Does this replace SEO, or is it just llms.txt?
Neither. The crawlers behind search and answer engines fetch raw HTML and never run JavaScript, so the pages still have to exist as real HTML that answers the question in its first line. That work comes first, and agent surfaces sit on top of it. And they are not llms.txt: consumer AI crawlers do not request that file, Google Search ignores it, and no study has shown a citation lift from publishing one. An MCP server and a JSON API are different in kind, because they are called on purpose by an assistant that is trying to answer a question right now, rather than left on the floor hoping a crawler picks them up.

For example: 136 KB social preview image, down from 1.74 MB. Hotel Marien

Book a callTell us about your project

Get started

Tell us what you want to build.

Websites, apps, custom systems and AI for companies that need them in production in weeks. Fill in the form or book a 30 minute call: a senior engineer answers within 48 business hours with a first take on scope, timeline and price.

Book a 30 minute call
What do you need?
Budget
When

At least ten characters.

Answer in writing within 48 business hours. No newsletter, no automatic sequence.

Rather answer the full brief?

First step

Discovery Sprint

Two weeks to close scope, architecture and price before anything is built.

No price list. You get a private quote, in writing, after a short call.

Then, to build, always at a fixed price

Ask for your quoteHow we quote

Not ready yet?

AI-native digital studio for mid-market companies

POLARGATE, S.L. · CIF B24895005
Calle Electrónica 6, 28027 Madrid, Spain