Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.0xarchive.io/llms.txt

Use this file to discover all available pages before exploring further.

Agent market-data tools should be chosen by job shape: generated code, hosted tool call, local Skill, CLI job, typed SDK, REST request, or WebSocket stream. The agent host is not the same thing as the tool surface. Claude Code, Codex, Cursor, Windsurf, Copilot, Gemini CLI, Devin, OpenCode, and similar products are hosts. OpenAPI, Markdown docs, CLI, SDKs, MCP Server, Skills, REST, and WebSocket are surfaces.

Comparison

SurfaceBest use
OpenAPIRoute, parameter, auth, and response truth
Markdown docsConcepts, venue taxonomy, examples, caveats, and page citations
RESTDirect product API calls
WebSocketStreams, replay, local state, and gap handling
SDKsApplication code and typed helpers
CLIShell, cron, CI, notebooks, and agent terminals
MCP ServerMCP-capable clients that need typed local tools
SkillSkill-capable agents or OpenClaw workflows

Recommendation

Start agents with OpenAPI for routes and docs Markdown for context. Add execution only after the agent has classified venue family, symbol, data family, and response shape.

Agent Task Checklist

Before a tool-using agent leaves docs-reading mode, give it a narrow task packet: target venue family, symbol or market slug, route or channel, auth source, expected response envelope, pagination plan, freshness check, and maximum request scope. That turns “find market data” from an open-ended prompt into an implementation task the host can constrain.

Decision Rules

Use OpenAPI when the agent must generate a request, choose parameters, or type a client. Use Markdown docs when the agent must explain venue taxonomy, data availability, credits, gaps, or when a route should not be used for a buyer’s question. Use llms.txt and llms-full.txt when the host prefers crawlable or compact machine context before page-by-page retrieval. Use REST for bounded requests such as one order-book snapshot, one trade window, one candle series, or one data-quality check. Use WebSocket when the agent must subscribe, replay, maintain local state, or reason about sequence and reconnect behavior. Use SDKs when the generated code will live in an application. Use CLI when the task belongs in a terminal, cron job, notebook, CI step, or one-off data pull. Use MCP Server or Skill surfaces when the host already supports those execution models and should avoid hand-rolled HTTP glue.

Safety Checks

Before letting an agent execute calls, require it to name the venue family, symbol format, route or channel, response shape, error handling, and rate-limit expectation. It should also explain why REST, WebSocket, CLI, SDK, MCP Server, or Skill is the right surface for that job. If it cannot separate a Hyperliquid core perp from Spot, HIP-3, HIP-4, or Lighter, keep it in docs-reading mode until the route is clear. For production workflows, keep secrets outside prompt transcripts and browser logs. Agents can draft code and commands, but API keys should come from the runtime environment or secret manager.
Last modified on May 18, 2026