How an AI assistant or coding agent can read this documentation: llms.txt, the Markdown version of every page, the documentation's MCP server and the OpenAPI definition — and which source to trust.
This documentation is written for people and for machines. If you use an AI assistant or a coding agent — Claude, ChatGPT, Cursor, your own agent — to build on the Spendesk API, point it at the entry points below rather than letting it guess from a web search.
This page is about reading the documentation. To let an assistant work with your Spendesk data, see Connect an AI assistant (MCP).
Entry points
| What | Where | Use it for |
|---|---|---|
| llms.txt | https://developer.spendesk.com/llms.txt | The index of every page, with a one-line description: guides, conventions, recipes, the API reference, the changelog. Start here. |
| Markdown pages | Any page URL followed by .md, for example https://developer.spendesk.com/reference/pagination.md | The text of a page without the site around it. llms.txt links to these. |
| MCP server for the documentation | https://developer.spendesk.com/mcp | Lets an assistant list, search and read the API's endpoints, with their parameters and schemas (list-endpoints, search-endpoints, get-endpoint). No Spendesk account or sign-in needed. |
| OpenAPI definition | https://developer.spendesk.com/openapi/spendesk-public-api.json | The machine-readable description of every endpoint: paths, parameters, schemas, scopes. Use it to generate client code. |
| Postman collection | Postman collection | Ready-made requests for the whole API, with the token handled for you. |
The documentation's MCP server also has an execute-request tool, which sends a request you give it. Use it with demo credentials while you build.
Which source to trust
When two sources seem to disagree, use this order:
- The OpenAPI definition and the endpoint pages generated from it, for paths, parameters, schemas and scopes.
- The guides and conventions on this site, for how endpoints work together and for behaviour the definition cannot express: pagination, errors, rate limits, authentication. The pages that end with "Last checked against the Spendesk demo environment on …" are replayed against the demo API every week.
- The MCP tool reference, for the tools of the Spendesk MCP server: it is generated from the server's own tool registry.
- The Spendesk Help Center, for steps inside the Spendesk app and inside AI assistants (setting up a connection, enabling write permissions).
Other websites, blog posts and older answers may describe an earlier state of the API or of the MCP server. If an assistant quotes a number of tools, a feature "coming soon" or a menu path you cannot find here, check it against this site.
Tips for agents
- Experimental means it can change. Endpoints and scopes named
experimentalneed access to be requested and may change — see Versioning & Deprecation. - Try on the demo first. Build and test against
https://public-api.demo.spendesk.combefore production — see Getting access. - Write requests are real. There is no dry-run mode: test anything that creates or changes data on the demo.
- Read the errors. Every error carries a
correlationId; give it to [email protected] when a call fails for no clear reason — see Error Handling.
Found something an assistant got wrong because of this documentation? Tell us at [email protected].