For agents

Reductivist treats agents as visitors. This page lists what is here for them and how bots are handled, so an operator does not have to guess.

Read the site

MCP server

A remote MCP server over Streamable HTTP, with no sign-in, at this address:

https://www.reductivist.com/mcp

Its description is at /.well-known/mcp.json. These are the tools it offers.

search_products
Lists products, optionally filtered by a word in the name or description. Returns name, price, stock and URL. It changes nothing on the site.
get_product
One product in full by slug, with price, stock, description and photos. It changes nothing on the site.
get_policies
Shipping, returns and refunds, payment and tax, orders and contact, as the store states them. It changes nothing on the site.
get_cart_link
Gives the product page link where the person adds the item to the cart and checks out. This tool never buys anything. It changes nothing on the site.

The server allows 60 read calls per minute per server instance. It has no write tools. Requests over a limit get an error result.

WebMCP

In browsers that support WebMCP, a page registers these tools for an agent working in the browser when it loads the script at /agent-tools.js. Tools that fill a form leave the send button to the visitor.

search_products
Lists products, optionally filtered by a word in the name or description. Returns name, price, stock and URL. It changes nothing on the site.
get_product
One product in full by slug, with price, stock, description and photos. It changes nothing on the site.
get_policies
Shipping, returns and refunds, payment and tax, orders and contact, as the store states them. It changes nothing on the site.
get_cart_link
Gives the product page link where the person adds the item to the cart and checks out. This tool never buys anything. It changes nothing on the site.

Bot policy

Crawlers and agents are welcome to read every public page, including crawlers that collect text for model training. The generated robots.txt names each one. Requests are blocked at the edge when they name a verified bot such as Googlebot or GPTBot but do not come from that bot's addresses, when they ask for known scanner paths such as /.env or /wp-admin, or when one address sends more than 30 requests in 10 seconds to /mcp, /api or form posts. Bots that cannot be verified, ClaudeBot and Google-Extended among them, are not blocked.

search, allowed

Classic search engines.

ai-search, allowed

AI search that cites pages.

ai-user, allowed

Agents acting for a person.

ai-training, allowed

Collects text for model training.

Contact for operators

To report a problem with how a bot is treated, or to ask for a higher rate limit, write to this address:

jon@reductivist.com