MKRST for agents and developers

A public, read-only REST API at https://api.mkrst.co/v1, described by an OpenAPI 3.1 spec. No key, no sign-up.

• Services: /v1/services and /v1/services/{slug}
• Pricing: /v1/pricing, our engagement tiers with “from” prices in AUD
• Work: /v1/work and /v1/work/{slug}, with keyword search
• Journal: /v1/journal and /v1/journal/{slug}
• FAQs and contact: /v1/faq, /v1/contact
• Site search and any page as Markdown: /v1/search, /v1/pages
• Enquiries: /v1/enquiries (see below)

Every error is JSON, { "error": { "code", "message", "hint" } }, with the right HTTP status. The data is read live from this website, so it matches what you see here. How access works is in auth.md.

The same data is available as an MCP server over Streamable HTTP, with no authentication:

https://api.mkrst.co/mcp

It has 11 tools: list_services, get_service, get_pricing, search_work, get_work, search_journal, get_faq, get_contact, search_site, read_page and submit_enquiry. Its server card describes it.

Claude. In Settings, open Connectors, choose “Add custom connector”, and paste https://api.mkrst.co/mcp. Leave authentication empty. In Claude Code, run:

claude mcp add --transport http mkrst https://api.mkrst.co/mcp

ChatGPT. Turn on developer mode in Settings, add a custom MCP connector, paste the same URL and choose no authentication.

The menus move around as these products change. The only thing you need is the URL.

An agent can send us a new-business enquiry, but only with the user’s consent. It works like this:

1. The agent calls submit_enquiry (or POST /v1/enquiries) with dry_run: true. Nothing is sent. It gets back exactly what would be sent.
2. The agent shows that to the person it is working for and asks if they want to send it.
3. Only after a clear yes does the agent send it again with dry_run: false and user_confirmed: true.

A real enquiry needs a name, what the person needs, the agent’s name, and an email or phone number. It creates a lead in our CRM and emails the studio. A person at MKRST replies to the user directly, not to the agent.

Every page has a Markdown version, with the page title, description, URL and last-updated date at the top.

• Add .md to a page address, for example mkrst.co/pricing.md or mkrst.co/index.md for the homepage.
• Or request any page with the header Accept: text/markdown.
• The homepage also has a short agent view at mkrst.co/?mode=agent.

llms.txt is the index of the site for agents. There are also indexes for work, services and the journal.

• /.well-known/api-catalog: API catalog (RFC 9727)
• /.well-known/ard.json and /.well-known/ai-catalog.json: the MCP server and API, listed for agent discovery
• /schemamap.xml: our schema.org data (organisation, services, work and journal) as JSON Lines feeds, following NLWeb’s Schema Feeds format

In a browser that supports WebMCP, this website also offers three tools to the agent in the page: get_pricing, search_work and start_enquiry. The last one only fills in our contact form. The person still reads it and presses Send.

• Reads: 60 requests per minute per IP address. Every response carries RateLimit-Limit, RateLimit-Remaining and RateLimit-Reset. On a 429, wait for Retry-After seconds.
• Dry runs: 30 per minute per IP address.
• Real enquiries: 5 per hour per IP address.
• Data is cached for up to 5 minutes.

Need more, or something that is missing? Email [email protected].

We publish a public repository with ready-made skills for your own agent: github.com/caseyluke/mkrst-agent-skills. It has two Agent Skills, “Brief a branding agency” and “Work with MKRST”, an AGENTS.md, an MCP config for the server above, and an Agent Plugin manifest. MIT licence.

Install the skills with npx skills add caseyluke/mkrst-agent-skills.