MCP Server — AI Agent Access to the AKADATA Catalogue

MCP means Model Context Protocol. It allows an AI agent, coding tool, browser agent, or automation client to connect to a service and ask for clearly defined information through known tools instead of scraping pages, guessing routes, or inventing answers.

We built the AKADATA MCP server because AI agents are now part of the web. If an agent wants to understand AKADATA, discover public pages, search the site, or read public content, it should not need to pretend to be a browser or scrape random HTML. It should have a clean way in. That is what this server provides.

AKADATA is built to be readable by humans, search engines, and AI systems.

The public website is rendered as proper HTML. The sitemap is available. The llms.txt file explains the site to language models. The /webmcp.json file publicly describes the available MCP-style tools and endpoints.

The MCP server completes that path.

It gives AI agents a proper interface to the same public data that users can browse on the site. It also lets us keep control of safety, authentication, rate limits, and access rules without hiding what the platform does.

We do not want agents scraping blindly. We do not want fake traffic pretending to be human. We do not want useful public catalogue data locked away behind unnecessary friction either. So we made a proper door.

Basic MCP access is free.

The public read-only MCP server can be used by an AI agent or MCP client without creating an account or API key.

That free access is enough for normal reading, testing, public site search, page discovery, and structured content retrieval.

It is free because that is how we do things. Useful public access should not be made difficult just because the client is an AI agent instead of a person with a browser.

The public MCP description is available for anyone to inspect at /webmcp.json. The tool list and current read-only endpoints are public.

AKADATA also publishes a small cookie-free OAuth service for optional protected agent credentials. Anonymous clients can register, receive a client secret once, and exchange it for a short-lived ES256 bearer token. Claims, introspection and revocation endpoints are publicly discoverable through /auth.md.

AIE remains the encrypted browser form layer. OAuth provides the interoperable agent credential layer. They have separate keys and separate responsibilities.

The current MCP tool level remains public and read-only. An agent can read public site content, but it cannot change accounts, submit forms, update content, or perform manager actions.

Merchant-specific MCP tools can come later.

The first version is deliberately simple: public catalogue and platform read access only.

In future, a merchant account may be able to create merchant-scoped API keys, allowing AI tools to read merchant-specific data or help manage merchant information through carefully limited permissions.

That will only be added when the permission model is correct.

The web is changing.

Search engines still matter. Human visitors still matter. But AI agents now read, summarise, compare, recommend, and automate.

AKADATA is being built so those agents can interact honestly.

We would rather publish clear rules, clear endpoints, and clear access paths than force agents into scraping behaviour that looks like abuse.

The MCP server is part of that principle.

AKADATA public read-only MCP access is free.

Connect an MCP-compatible agent using the public manifest and endpoint. No account or API key is required for the current read-only tools.

Public tool description: /webmcp.json. Optional protected agent registration and bearer-token instructions are published at /auth.md.