Skip to main content
These docs are written to be read by agents as much as by people. Everything here is available as plain text, and as an MCP server your coding agent can search.

Give your agent these docs

llms.txt

An index of every page with a one-line summary. https://docs.orbb.com/llms.txt

llms-full.txt

The whole site in one file, for an agent that wants everything in context. https://docs.orbb.com/llms-full.txt
  • Any page as Markdown: add .md to its URL — https://docs.orbb.com/quickstart.md.
  • The menu on every page copies it as Markdown, or opens it in Claude, ChatGPT or Cursor.
  • The API as a spec: the OpenAPI 3.1 document, with every field and error code, is linked from the API reference.

The docs MCP server

Install the docs themselves as an MCP server, and your coding agent can search them while it writes your integration:
This is a different server from Orbb’s own MCP: it searches these pages, it does not call Orbb.

Rules for an agent writing against the API

Paste these into your agent’s instructions, or let it read this page.
1

Branch on error, never on detail

error is a stable code. detail is a sentence for a person and its wording changes. See Errors.
2

Handle 202 on every job call

Resolve, circle, edge and paths can answer 202 with a job. Poll /v1/job with the same id — every 3 seconds for 10 polls, then every 10. A client timeout under 60 seconds loses the job id, and a retry pays again. See Jobs.
3

Never read an absent value as zero or no

A missing strength is not 0. An UNKNOWN verdict is not NO_EVIDENCE_FOUND. Report coverage.not_examined alongside what was found. See Unknown is not no.
4

Never pick a candidate silently

AMBIGUOUS resolves and 422 ambiguous_person come with candidates. Choose with real context, or ask.
5

Respect Retry-After

On any 429, wait Retry-After seconds plus a little jitter, then repeat the same request. Keep at most 10 calls in flight, polls included. See Rate limits.
6

Do not research by default

on_miss: RESEARCH costs extra and runs longer. Call with the default first, then research only the gaps that change the answer.