| name | ot-ref |
| description | Use when calling any OneTool pack tool via __onetool/__ot triggers or the MCP run tool — pack map with aliases, call syntax, kwarg-prefix and alias forgiveness, discovery, a greppable index of every command, recovery from disconnected servers, and large-result handling. |
| tags | ["reference","tools"] |
OneTool Reference
Load this whenever you work with OneTool tools — ideally before your first call.
OneTool exposes 240+ tools across ~28 packs through one MCP tool: run(command='pack.tool(arg=value)').
Call conventions
Two request forms:
- Literal call:
__ot ground.search(q='mcp features 2026') → pass through and execute exactly as written.
- Natural-language intent:
use __ot ground.search to see MCP features 2026 → map to ground.search(query='mcp features 2026'), synthesizing args from the stated goal; inspect the signature first if unsure.
Rules:
- Call shape is
pack.tool(arg=value) — never ot.pack.tool(...).
- The
: prefix is snippet syntax only (:pkg_npm packages=react); direct calls never take a colon.
- Use keyword args for keyword-only tools; prefer single-quoted string literals.
- Python glue is allowed (variables, dict/list transforms, last-expression returns). Arbitrary imports are blocked — use pack tools instead.
The engine is forgiving — rely on it
- Kwarg prefixes: a kwarg resolves to any signature param it prefixes —
q= → query=, pat= → pattern=. Exact match wins; an ambiguous or colliding prefix errors instead of guessing.
- Pack aliases: verbose packs have short aliases (pack map below) —
wb.draw(...) ≡ whiteboard.draw(...).
- Proxied MCP tool names match in snake/camel/Pascal case:
github.list_repositories ≡ github.listRepositories.
Pack map
- arch — Architecture workflows for Excel ingest, validation, generation, round-trip, and bundling
- brave (br) — Search the web, news, and images — fast, private, with batch support
- chrome_util (chrome) — Annotate and highlight page elements via Chrome DevTools
- console — Publish inline text, markdown, code, diffs, and more to a connected onetool-console app
- context7 (c7) — Pull up-to-date docs for any library — React, FastAPI, etc.
- convert (cv) — Turn PDF, Word, Excel, and PowerPoint files into clean markdown
- db — Explore and query databases — list tables, inspect schema, run SQL
- diagram (diag) — Create diagrams — Mermaid, Graphviz, PlantUML, D2, Excalidraw [experimental]
- excel (xls) — Work with Excel files — read, write, search, pivot, and set formulas
- file (f) — Read, write, edit, copy, move, and delete files; browse directory trees
- ground (g) — Research any topic with Google AI-grounded search and source citations
- knowledge (kb) — Portable SQLite knowledge bases with hybrid FTS5+vector search and AI synthesis
- localhist (lh) — Project-local snapshot history — save, list, diff, and restore file versions
- mem — Persistent topic-based memory — write, search, grep, slice, and organise
- ot — ** Discover tools, packs. Get the help you need to call them correctly. The BEST PLACE to start. **
- ot_context (ctx) — Smart context store — ask, read, search, grep large tool responses without filling context window; format-aware toc/slice/query for json, yaml, and markdown
- ot_forge (forge) — Scaffold and generate code from templates
- ot_image (img) — Vision analysis via dedicated model — load images, ask questions, get summaries; zero tokens to host session
- ot_llm (llm) — Run any text through an LLM — summarise, extract, reformat, translate
- ot_secrets (sec) — Store and retrieve API keys and secrets for OneTool services
- ot_servers (srv) — Runtime proxy server controls — enable, disable, restart, and inspect one server
- ot_timer (tmr) — Time tool calls and report performance
- package (pkg) — Look up PyPI and npm packages, audit deps, find the latest AI models
- play_util (play) — Annotate and highlight page elements via Playwright
- ripgrep (rg) — Search code and text at speed — regex, globs, file types, context lines
- tavily (tav) — Search the web with Tavily AI — clean results, answer summaries, URL extraction
- webfetch (wf) — Fetch any URL and extract clean, readable content
- whiteboard (wb, excalidraw) — Live whiteboard using Excalidraw — you can draw diagrams with a powerful DSL, assisted by AI
Proxied MCP servers appear as additional packs — ot.servers() lists them.
Find any command
The full signature index ships with this skill: reference/tool-index.md — one line per tool in
the form pack.tool(args) # description, grouped under ## pack, alias headings.
Grep it with your file tools; never read the whole file into context:
rg -n 'pivot' reference/tool-index.md
rg -n '^whiteboard\.' reference/tool-index.md
Runtime equivalents when calling beats grepping:
ot.tools(pattern='brave.', info='signatures') — the same one-liner format for one pack (~200 tokens).
ot.tool_info(name='pack.tool') — full detail for one tool.
ot.help(query='topic') / ot.help(ask='how do I ...?') — semantic discovery.
Recovery
- Disconnected proxy server:
ot_servers.enable(name='playwright') then retry once; ot.servers() first if name/status unknown.
- Unknown or typo'd tool:
ot.tool_info(name='pack.tool') suggests corrections; ot.packs() lists packs.
- Failed call: inspect once with
ot.tool_info, repair kwargs, retry once. Do not guess beyond one retry.
- Deep dive (proxy handling, ctx navigation, output dunders, run-vs-local-script):
reference/recovery.md.
Large results
Oversized outputs are stored and returned as a handle: {'handle': 'b2d18a1b', ...}.
- Always pass the handle string, not the dict:
ot.result(handle=h['handle'], search='error').
ot.result(handle=..., offset=, limit=, search=, tail=) works on every install. With the
[util] extra, the richer ctx pack adds ctx.toc/read/slice/grep/query/ask — see
reference/recovery.md.
Output controls
Set as a statement before the call: __format__ = 'yml_h'; ot.help(query='search').
Dunders: __format__ (json, json_h, yml, yml_h, raw), __sanitize__,
__force_context__. Details in reference/recovery.md.