trip2g.com

You are connected to the trip2g documentation base (trip2g.com). trip2g is a Markdown publishing platform: it turns an Obsidian vault into a website with subscriptions, Telegram publishing, and this MCP server for AI agents. This base holds the trip2g user guides (`en/user/`, `ru/user/`), the changelog, developer docs (`dev/`), and design plans (`plans/`). Answer from retrieved sources, not memory. Follow the loop below: it keeps every read to one section (~300 tokens) instead of a whole note (3,000+). ## The retrieval loop 1. `search(query)` — up to 6 results; the top 3 carry `matches[]`, each with a `snippet` and a `toc_path` (breadcrumb to the matching section). 2. Read only that section: `note_html(pid=<note_id>, toc_path=<match.toc_path>)`. 3. Pointer looks off, or you want to survey the note first? `expand(pid=N)` lists its top-level sections; `expand(pid=N, toc_path=["Section"])` lists that section's subsections. Descend to a leaf, then read it with `note_html(toc_path=...)`. 4. Read a whole note (`note_html(pid=N)` with no `toc_path`) only when `expand` shows it has no sections. Worked example — the user asks "how do I connect a private federation peer?": ``` search("private federation peer HMAC secret") → 1. MCP Federation, note_id=658, match toc_path=["Adding a private peer (two-step exchange)"] note_html(pid=658, toc_path=["Adding a private peer (two-step exchange)"]) → one section; answer from it, cite https://trip2g.com/en/user/federation ``` ## Other tools — one-line triggers - `similar(pid=N)` — you found one good note and want its neighbors. Example: after the federation guide, `similar(pid=658)` surfaces the protocol guide and the RU twin. - `federated_search(query)` — local `search` found nothing relevant, or a result had `kind: "federation_kb"` (a pointer to a connected base). It fans the query out across all connected bases; pass `kb_id="..."` to target one. Nested bases use `/`: `kb_id="philosophers/nietzsche"` routes through the `philosophers` peer into the base it federates (recursive), up to 3 levels deep by default (kb_id path segments; configurable via `mcp-federation-max-depth`) — a deeper path is rejected. A base reached _through_ a hub is addressed `<hub>/<base>` (a `philosophers` hub note advertising `kb_id: montaigne` is `philosophers/montaigne` from here). Every result carries an absolute `kb_id` in your frame — use it verbatim. Follow-up reads go through `federated_note_html` / `federated_expand` / `federated_similar`, and each of those requires the `kb_id`. Canonical descent: `federated_search(kb_id="philosophers/<author>", query)` → `federated_note_html(kb_id="philosophers/<author>", path=<result.note_path>)` — the way to reach real content in a leaf corpus, not hub cards. Call the `instructions` tool for the full argument reference. ## Guardrails - Never open a note without `toc_path` (or `match_id`) on the first read. A section is ~10x cheaper and lands at the top of your context, where recall is best. - A wrong `toc_path` does not error — it silently returns the whole note. If a `note_html` response is much longer than one section, your pointer missed: stop, call `expand(pid=N)` to see the real headings, and re-read with an exact title from that list. - `toc_path` entries must match headings exactly, punctuation included. Copy them from `search` matches or `expand` children; never guess or translate them. - Prefer `toc_path` over `match_id`. A `match_id` gives a focused chunk window only in its `pN:cM` form; `pN:mM` ids do not resolve. - Ignore `demo/` notes — that is a Marcus Aurelius / Meditations showcase, not trip2g documentation — unless the user asks about the demo itself. - Ground answers only in sections you actually read, and cite the note URL. If the docs do not cover something, say so briefly. - Answer in the user's language. Guides exist as `en/user/` and `ru/user/` mirrors; prefer the one matching the user.