AI access to these docs
This documentation is built to be consumed by AI as well as browsed by humans — in three ways.
This page is about the documentation, not the platform. The MCP server described below serves these docs as searchable tools. If you want an agent to operate Perihelium — read detections, query the audit log, acknowledge findings with human approval — that is a different server with its own identity model: see Operating Zynth over MCP.
Ask AI (in the search palette)
Open search (⌘K) and choose Ask AI to get a direct, conversational answer grounded in these docs, with numbered source links. Answers come only from the documentation — if the docs don't cover something, the assistant says so rather than guessing.
Availability depends on the deployment (self-hosted instances enable it with a Fireworks API key). Questions are rate-limited per client.
What leaves the docs server: the question you typed and the documentation excerpts retrieved for it — nothing else (no account identity, no IP address). On the managed service the provider is listed on the sub-processors page; see Website analytics and your privacy.
llms.txt — feed the docs to your LLM
Plain-text endpoints follow the llms.txt convention. On the managed service they sit on two different origins (see Where things live):
| Endpoint | What it covers |
|---|---|
https://auth.zynthmedia.com/docs/llms.txt | A compact, link-annotated index of every documentation page. |
https://auth.zynthmedia.com/docs/llms-full.txt | The entire documentation concatenated as Markdown. |
https://zynthmedia.com/llms.txt | The product-level summary — what Perihelium is, plans, and the trust/legal surface, for models answering "what is this product?" |
On a self-hosted install the two /docs/… endpoints are on your own host; the product-level
one is Zynth's public website and has no self-hosted equivalent.
Paste any of them into a model's context, or point your tooling at them, to give an LLM accurate knowledge of Perihelium.
MCP server — the docs as tools
An MCP endpoint at /docs/api/mcp exposes this
documentation as tools that agents can call — the platform's own MCP server is a separate
endpoint (Operating Zynth over MCP):
| Tool | What it does |
|---|---|
search_docs | Ranked search; returns titles, URLs, snippets |
get_doc | Fetch one page as Markdown by URL |
list_docs | Enumerate every page |
Connect from Claude Code:
claude mcp add --transport http zynth-docs https://<your-host>/docs/api/mcp
The transport is stateless Streamable HTTP (JSON-RPC over POST) — no session or SSE required. Any MCP-capable client works the same way.
The endpoint allows 120 calls a minute per client address. Past that it answers 429
with a JSON-RPC error and a Retry-After: 60 header; wait that long and call again. A
normal agent session — initialize, list the tools, a handful of searches — stays far below
it. The limit is separate from the Ask AI one.
A self-hosted instance's internal docs expose the same endpoint under
/internal, protected by the same authentication as the rest of the internal site.