AI & agents

Point Claude, Cursor, ChatGPT, or your own agent at these docs — and integrate against Drazill correctly on the first try.

These docs are built to be read by machines as well as people. Every page is generated from the same CI-gated contract the exchange runs on, so what an agent reads here cannot drift from the code that serves its requests. This page is the map of the machine-readable surface — and the safety rules an agent should carry into any Drazill integration.

Drazill settles real money. Whatever tool or agent you use, the rule is the same: never put a drzl_live_ key in a browser, a notebook, or any client you don’t control. Build and test against staging (https://staging.drazill.com) with drzl_test_ paper keys first. The AI assistant on this site is configured to enforce exactly this in every answer.

Page actions — copy, Markdown, and one-click agent hand-off

Every page carries a page-actions menu (top-right) with:

  • Copy page / View as Markdown — grab the current page as clean Markdown to paste into any model’s context.
  • Ask AI — open the on-site assistant (see Ask Fern below).
  • ChatGPT / Claude — open the page in that assistant with the content preloaded.
  • Claude Code / Cursor — open the page and register this site’s documentation MCP server in your editor, so the agent can search the whole docs set, not just the page you’re on.
  • VS Code — open the page’s Markdown in your editor.
  • Download OpenAPI spec — a direct link to the pinned, versioned contract (GET https://api.drazill.com/api/v1/openapi/v1.json, ETag-cached). Generate a client in any language from it.

The documentation MCP server

This site hosts a Model Context Protocol server at:

https://drazill.docs.buildwithfern.com/_mcp/server

Add it to Claude Code, Cursor, or any MCP-capable client (the Claude Code and Cursor page actions register it for you) and your agent can search and cite these docs while it works.

What this MCP server is — and isn’t. It is a documentation server: it searches and answers questions about the docs (endpoints, concepts, error codes, webhook shapes). It does not execute Drazill API calls, hold keys, or move money — it has no access to your account. A hosted API MCP server (one that calls Drazill endpoints on your behalf) is not offered today; if that changes it will be documented here and announced in the changelog. Until then, agents call the API the normal way — over HTTP with an X-API-Key header, using the reference this server points them at.

llms.txt and per-page Markdown

For agents that prefer to pull context directly rather than through MCP:

  • https://drazill.docs.buildwithfern.com/llms.txt — an index of the documentation in the llms.txt format, auto-generated by Fern. Nested sections expose their own scoped index (for example …/webhooks/llms.txt), so an agent can load just the slice it needs.
  • Per-page Markdown — every page is available as raw Markdown via the View as Markdown / Copy page actions above. This is the honest, context-window-friendly way to feed a specific page to a model.

There is no llms-full.txt on this site. Fern deliberately does not concatenate the whole site into one file — for a reference this size it would blow past most model context windows and be of little use. Point your agent at the scoped llms.txt for the area it cares about, or hand it individual pages’ Markdown, instead.

Ask the docs (Ask Fern)

The Ask AI action opens an on-site assistant that answers from these pages with citations. It is configured with Drazill’s safety mandate baked into its system prompt — it will steer you to test keys and staging, warn on money-moving endpoints, and refuse to invent behaviour that isn’t documented. Reader queries are PII-masked before processing.

Ask Fern runs on the docs’ free Hobby plan, which meters AI usage at 250 credits per month. If that proves tight we’ll measure real usage before deciding on more — we never silently upgrade. See the owner runbook for the monthly credit check.

Privacy & data on this documentation site

This site is public, marketing-class documentation. It is a separate property from the Drazill app and is bound by the same truth-first, privacy-respecting rules:

  • No forms, no accounts, no PII capture. The docs collect no personal data and have no login. The ca-central-1 data-residency rule governs user data, which this site never holds.
  • Analytics. Docs analytics are not enabled today. If they are turned on, the recorded plan is PostHog pageview analytics only, with no session recording — the vendor and data region are an owner decision recorded in the runbook, and nothing ships until it is ratified. What would be collected (page, referrer, coarse device class — no emails, names, or precise location) is stated here before any of it is switched on.
  • Ask Fern queries are processed by Fern’s AI search with PII masking on; do not paste secrets (like a real API key) into the assistant.

Start here

Quickstart

First call in under a minute — unauthenticated market data, then a paper trade.

Authentication

API keys, key classes, scopes, and why live keys never belong in a browser.

API Reference

Every REST operation, rendered from the OpenAPI contract. Download the spec above.

Errors

The full error catalog with remediation — grounded in the enforcement code.