autario developer documentation
autario is data app infrastructure. You bring the idea, autario brings roughly eight thousand normalized public datasets, the private data your users connect, a sandboxed runtime, and the URL your app ends up on. These pages are written so that a developer who has never spoken to us can go from an empty editor to a published app without asking a single question.
Start here
If you have ten minutes and an LLM client, read the quickstart. It walks the same lifecycle three times, once for each way in: through MCP with a chat assistant driving the calls, through REST with curl, and through the Builder by clicking. All three end at the same place, a published app with a shareable URL.
- Quickstart: your first app in ten minutes: the three ways in, end to end
- Authentication: API keys for REST, OAuth for remote MCP, scoped tokens for apps
- Reading data: search, query, aggregate, precomputed statistics
- Writing data: create a dataset, append rows, keep it fresh
- Connectors: hosted, refreshing tables from vendor APIs
- Apps: create, preview, publish: the visibility states and what the curator still decides
- autario.js: the one-line data SDK every app gets
- MCP tool reference: every tool the server exposes, generated from the server itself
- Limits and pricing: generated from the live plan catalog
- Errors and refusals: error codes, the 402 contract, retries
- Change policy: dataset structure changes, API versioning, MCP releases
What you can build
An app on autario is a small self-contained web bundle. It runs in a locked sandbox with no outbound network of its own and reads data through a server-side bridge that enforces the scope you declared in your manifest. That constraint is the product: a user can open a stranger's app against their own connected data without auditing the code, because the code never holds a credential and never reaches the internet.
- Charts and widgets that render one verified series and stay correct because the number comes from the source, not from a copy.
- Analyses that join several datasets on entity code and time, which autario normalizes at ingestion so you do not write the mapping.
- Tools that run on a user's own connected data, where the same app serves every customer and never sees a key.
- Reports that save their output as an artifact, so the result has a URL and can be reopened, embedded or cited later.
There is nothing to install to try it. The public half of the data API answers anonymous requests, the reference app source is fetchable, and a free account is only needed once you want to save something.
The three documents a machine should read
These pages are for humans and for crawlers. An agent that is going to call autario should read the machine-facing set instead, which is denser and always current.
- Agent guide: the whole surface as one compact text payload, written for a system prompt
- OpenAPI document: every documented REST path with its request and response schemas
- llms.txt: the map an assistant reads first, with the token arithmetic that explains why to call us
- Sandbox runtime contract: the tier-2 iframe, the bridge ops and the CSP, in machine form. It registers a manifest through the older public door, so follow the quickstart above for the current lifecycle
- autario.js: the data SDK inlined into every app, one file, zero dependencies