Developer tooling for the rConfig API
rConfig describes its own API in several forms, each for a different reader: a person browsing endpoints, a tool such as Postman, or an AI agent writing code. They are all generated from your own install, so they match the API your server actually has. This page explains which to use when, and how to get an AI coding agent to write a sound integration.
Which one to use
| You want to | Use | Where it is |
|---|---|---|
| Browse endpoints, parameters and examples | In-app API docs | Settings → REST API → API Documentation in rConfig |
| Import the API into Postman, Insomnia or Swagger UI, or generate a client library | OpenAPI file | Generated on the server; see OpenAPI specification |
| Have an AI coding agent write code against the API | AI-agent REST API docs | AI Center → API Docs; see AI-friendly API docs |
| Have an AI agent look things up in rConfig, following recommended workflows | rConfig Skill plus the MCP server | AI Center → Skills; see rConfig Skill |
The first three describe the REST API. The Skill is different: it teaches an agent to use rConfig’s MCP server, not the REST API. That belongs to the Iris guides, not to integration code.
Where the descriptions come from
The OpenAPI file and the AI-agent docs come from one generation run on your server, built from the API’s routes and the validation rules behind each endpoint. Two things follow:
- They describe the API as it was when they were generated. Regenerate them after every rConfig upgrade, or new and changed endpoints won’t be in them.
- They describe requests better than responses. Generation reads the code; it never calls the live API. Request fields come with their types and limits (a device name that must be at least 5 characters says so), but many responses have no recorded example. Send one real request to see the actual shape before writing code that parses it.
The OpenAPI file is written to the server and never served over a URL, so copy it off the server to use it. The OpenAPI page has the steps.
Generating a client
An OpenAPI generator can turn the file into a typed client for most languages. Worth knowing first:
- There are two files, one for v1 and one for v2. Most integrations need both, because device create and update, downloads and snippets are v1 only. See v1 or v2.
- The security scheme is an API key in the
apitokenheader. Generated clients pick that up from the file; don’t switch them to bearer authentication. - Responses are loosely typed because of the missing response examples. Expect to add your own types for the responses you use.
For a few calls, a hand-written client like the ones in pick a language is often less work than a generated one.
Writing integrations with an AI coding agent
AI coding agents (Claude Code, Cursor, GitHub Copilot and others) write good API clients when they have an accurate description of the API. Give them the AI-agent REST API docs rather than letting them guess from memory.
- Generate the AI-agent docs in AI Center → API Docs (there’s one file for v1 and one for v2), and give both to the agent: attach them, or save them in the project so the agent can read them.
- Tell the agent the rules this API has: the token goes in the
apitokenheader, v2 lists are paged and capped at 100,download-nowonly starts a backup. The how the API behaves page is a good thing to point it at too. - Give the code a token scoped to the job, never your own. The code the agent writes is limited by that token’s abilities, whatever it tries.
- Run it against a non-production rConfig first, and read what it does before you schedule it.