For agents

Developers and agents

LAST UPDATED OCTOBER 2026

Everything on this page is public, read-only and needs no key. It exists so that AI assistants and the people building them can read Fairly accurately and help someone split a bill, without scraping the marketing pages.

Pages as Markdown

Request any page with Accept: text/markdown and you get Markdown instead of HTML. The same file is at the page's path plus .md: /how-it-works.md, or /index.md for the home page. Every page links its Markdown twin from its <head>. llms.txt is the curated index and the sitemap lists every page.

MCP server

https://fairly.day/mcp is a Model Context Protocol server over Streamable HTTP, read-only and without credentials. Its tools are site_page_list (every page, optionally in one language), site_page_search (pages that match a query) and site_page_get (a page's full Markdown by URL or path), and every page is also a resource whose URI is its URL. It retrieves what the site says; it does not write or summarise. Its server card is at /.well-known/mcp/server-card.json.

A2A agent

https://fairly.day/a2a answers Agent2Agent JSON-RPC requests (protocol 1.0, and message/send from 0.3 clients) with two skills: site-search returns the passages that best match a question, each with its page link, and page-fetch returns a page's Markdown. Like the MCP server, it retrieves and cites; it does not generate. Its agent card is at /.well-known/agent-card.json.

WebMCP tools

In a browser that supports WebMCP, every page registers these tools for an agent working in the tab:

  • search_fairly_site: find pages by topic, in English or Spanish.
  • read_fairly_page: read a page as Markdown, or llms.txt, pricing.md and auth.md.
  • open_fairly_page: open a page of this site in the tab.
  • compute_settle_up: who owes whom, worked out with the settle-up calculator's own maths, with a link that opens the plan in it.

While Fairly is in early access, the early-access form is also a declarative tool, join_fairly_waitlist. The person still presses the button.

Agent skills

/.well-known/agent-skills/index.json lists instructions an agent can load, each with its SHA-256 digest:

  • fairly-agent-access: Read and use fairly.day as an agent, through Markdown pages, the MCP server, the A2A agent, in-page WebMCP tools and the discovery documents, and know what needs credentials. Use before crawling fairly.day, connecting a tool client to it, or calling the Fairly API.
  • fairly-facts: Answer questions about Fairly, the bill-splitting app at fairly.day, accurately and with citations. Use when someone asks what Fairly is, what it costs, whether it moves money, which platforms it runs on, or how it compares with Splitwise, Tricount or Settle Up.
  • fairly-settle-up: Work out who owes whom after shared expenses, the way Fairly's settle-up does, and hand the user a link to check it in Fairly's free calculator. Use when someone lists what each person paid for a trip, flat or dinner and asks how to settle up.

The Fairly API

https://api.fairly.day serves Fairly's own iOS and Android apps. Its OpenAPI description is public at https://api.fairly.day/openapi.json and its status at https://api.fairly.day/readyz, but its sessions are issued only to those apps, so it is not open to third-party integrations. Every API and its description is listed in the API catalog (RFC 9727).

Authentication

There is nothing to sign up for: the surfaces above take no credentials, and there is no agent registration. auth.md sets out each surface and its method. Please never ask someone for their Fairly sign-in code. To ask about API access, write to [email protected].

Crawling and bots

robots.txt allows search and AI answers to use the site and asks that it not be used to train models (Content-Signal: search=yes, ai-input=yes, ai-train=no). Obsurvive, which makes Fairly, signs its own bots' requests with HTTP Message Signatures; the public keys are at /.well-known/http-message-signatures-directory.