load_onfire_guide
The MCP server ships long-form playbooks that are too large to live inside
a tool description. load_onfire_guide fetches one on demand.
When to use it
Section titled “When to use it”- Once per conversation, before your first use of a covered tool, when the task is more than a straightforward call — a distribution, a trend, evidence tracing, chaining datasets.
- After a validation error you don’t immediately understand.
A simple lookup doesn’t need a guide: the tools validate server-side and hand back the exact fix for free. Guides are static, so a second load in the same conversation is pure waste.
Inputs
Section titled “Inputs”| Parameter | Type | Required | Description |
|---|---|---|---|
guide_name | string | Yes | One of the guide names below. |
Available guides
Section titled “Available guides”| Guide | Covers |
|---|---|
ask_onfire | QueryIR authoring: entity choice, the provenance recipe (“how do we know X”), aggregate mode, worked examples, field-name traps. |
ai_prospecting | When (not) to run prospecting, the run/poll protocol, and how to read top_picks / priority_summary / the dataset. |
datasets | Dataset handles: from_dataset filters, query_datasets SQL slicing, and why dataset ids never travel through context. |
An unknown name returns an error plus the list of available guides.
Returns
Section titled “Returns”The guide body as markdown text.
Notes & limits
Section titled “Notes & limits”- Not tenant-gated: the guides are static documentation about the tool surface — no tenant data, no billing — so they are reachable for every tenant regardless of the tool allowlist.
- Byte-stable within a server version.