Tools
Every MCP tool Sprout exposes, grouped by the noun it touches. One page per noun, plus LLM bundles for direct agent consumption.
Every MCP tool Sprout exposes, grouped by the noun it touches. One page per noun. For the prose-flavored "what each noun is for," see Model: The nouns; this section is the dense lookup layer. For the push side of the protocol, see Subscriptions.
For your LLM
If you'd rather hand the docs to your own model directly, two bundles are served as plain text:
llms.txt: a structured index of every docs page with one-line summaries. Drop into a system prompt.llms-full.txt: the full docs corpus concatenated as plain text. Use when you want your model to read everything in one pass.
Both follow the emerging llmstxt.org convention; they regenerate on every docs build.
The catalog
Conventions
Every tool follows the same shape on the wire:
- One required scope (sometimes a second high-sensitivity one). See Scopes and sensitivity tiers.
- An input schema; the server validates and returns
BAD_INPUTon shape violations with a directional error message. - A return shape documented on each tool's section.
- One of four error codes:
BAD_INPUT,PERMISSION_DENIED,DOMAIN_NOT_FOUND,INTERNAL_ERROR. See the error envelope for what each means. - Writes are safe to retry. Send a stable
Idempotency-Keywith the call (HTTP header on the MCP transport, or the equivalent field your client exposes); the server dedupes and returns the cached envelope.
What's documented here is the partner-facing wire surface. Sprout reserves the right to add fields, codes, and read-shape extensions; breaking changes go through a versioned migration.