ourword-mcp
This server lets an AI model look up a curated library of 403 classic texts by situation, not by keyword, and return source-linked passages.
browse(): list ~121 situation groups (English) or ~112 (Chinese), or drill into a group to see exact wordings and matching chapters.read_chapter(url): fetch the full text of a chapter using a URL returned by browse/search.search(query, limit): literal keyword fallback when browse fails, limited to 10 results.Supports Chinese (
zh, default) and English (en) indexes, which are not translations of each other.Returns only what is actually in the library; refuses to invent matches.
Every result carries a URL so answers can point back to the source text.
Includes disclaimers that content is historical perspective, not personal medical, legal, or financial advice.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ourword-mcp I keep procrastinating on important work — what chapter speaks to that?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ourword-mcp
403 deep reads on 171 figures and classics across 2,600 years, indexed by the situation you're actually in. An MCP server with no search — the client's model does the matching.
Why there is no search in here
Someone says: "my boss keeps changing priorities."
The entry in the library that answers it reads: "Five things are on fire and I am spread across all five."
Those two sentences share zero words. Keyword search cannot get from one to the other; the model calling this server can, instantly. So the server stopped trying. It exposes the index in two levels and lets the model pick:
browse() → 121 situation groups
browse(group="Everything is urgent…") → the specific wordings, each with chapters
read_chapter(url) → full textThat design is not a preference. The first version was a literal search: on real situation sentences it returned 0 hits, top score 0.11. The scoring function had its denominator rewritten twice before the conclusion became obvious — nothing belonged in that position at all.
search() still exists as a fallback for when the model cannot pick. It is
documented as a fallback, in the tool description the model reads.
Related MCP server: BookLab MCP Server
Install
claude mcp add ourword -- uvx ourword-mcpBefore that lands on PyPI, straight from the repo — same one line:
claude mcp add ourword -- uvx --from git+https://github.com/woowoeth/ourword-mcp ourword-mcpOr from source, no packaging involved — it is one file and the standard library:
claude mcp add ourword -- python3 /path/to/ourword_mcp.pyNothing to clone: the index is fetched from the web and cached locally for a day.
The three tools
tool | what it does |
| no |
| full text of one chapter. Use the url |
| literal fallback. Prefer |
lang is zh (default) or en.
Chinese and English are not translations of each other
The same 403 chapters, entered through a different set of doors: 112 situation groups / 655 wordings in Chinese, 121 / 629 in English. What an English speaker says to themselves at 2am is not a translation of what a Chinese speaker says — so the two indexes grew separately. The Chinese side has situations about social-insurance gaps and the graduate exam; the English side has its own.
What it will not do
Written into the tool descriptions, because that is what the model reads:
Only what is in the library. Nothing matches → say so. Do not invent.
Every item carries a URL. An answer that cannot point back to the source does not get written.
Not personal medical, legal or financial advice. How people before you thought about such questions, yes. What you should do, no.
Ask it about a cat losing fur and it returns found: 0 and tells the model, in
so many words, not to make something up.
Also available as an Agent Skill
npx skills add woowoeth/ourword-skills — for clients without MCP. It reads the
same index over plain HTTP (https://ourword.ai/en/llms.txt), so it works with
no configuration at all. MCP is for people who can wire up a server; the Skill
is for everyone else.
Privacy
No telemetry. The server fetches two public JSON files and caches them in
~/.cache/ourword-mcp. Install counts come from public CDN and PyPI numbers,
not from anything this program reports.
MIT. Source: https://github.com/woowoeth/woowoeth.github.io/tree/main/tools/mcp
Available Tools
3 toolsbrowseA
主入口。 不带参数:列出库里覆盖的全部处境归类(一百多组)。带 group:列出那一组下面的具体说法,以及每条对应的「以前的人怎么处理」。用法是你先读归类、按语义挑最像用户此刻的那一组 —— 字面往往对不上(用户说「领导总改优先级」,库里那条是「全都重要,我砍哪个都疼」)。只返回库里真有的内容;检索不到就说没有,不要编。每条都带 URL,答案要能指回原文。这里给的是「以前的人在同一处境里怎么处理」,不是医疗、法律或金融的个人建议。
Inputs: lang, group.
read_chapterA
取某一篇的完整正文。url 用 browse / search 返回的那个,不要自己拼。
Inputs: url (required), lang.
searchA
兜底的字面检索,只在 browse 挑不出来时用。处境是语义的,字面常常对不上。
Inputs: query (required), lang, limit.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
Each tool has a clearly distinct role: browse navigates by semantic category, search is an explicit literal fallback for when browse fails, and read_chapter retrieves full text by URL. The descriptions even specify usage order, so an agent is unlikely to misselect.
All names are lowercase imperative verbs and clearly indicate actions. The only minor inconsistency is that read_chapter follows a verb_noun pattern while browse and search are bare verbs, but this is readable and does not create confusion.
Three tools is well-scoped for this focused read-only reference server. Each tool earns its place: semantic browsing, literal search fallback, and full-text retrieval cover the essential workflow without unnecessary bloat.
The server's stated purpose is retrieving curated situational wisdom, and the tools cover the complete read-only journey: list/select categories, inspect entries, retrieve full chapters, and fall back to literal search. No create/update/delete operations are expected for this domain.
Maintenance
Related MCP Connectors
Search and cite passages from a vetted, rights-cleared reference library.
Search a catalog of 1.7 million books and find books by mood, vibe, pace or genre. Read-only.
Search and read saved employee handbooks with scoped, read-only access and policy source links.
AI-curated book catalog that eliminates hallucinations and surfaces lesser-known titles.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables AI assistants to help users manage their reading experience by searching books, tracking reading progress, managing bookmarks, and generating personalized recommendations and summaries.-
- AlicenseNot gradedqualityDmaintenanceProvides curated nonfiction book recommendations and profiles using TF-IDF scoring with concept expansion, enabling users to get ranked book suggestions by describing a situation, topic, or question.10 npmMIT
- AlicenseBqualityDmaintenanceEnables a shared reading room where users and Claude can import EPUB or text books, read chunk-by-chunk, annotate passages, search, track progress, and resume from unread sections.26MIT
- FlicenseNot gradedqualityNot gradedmaintenanceEnables AI agents to search, browse, and cite articles from a curated reading list through read-only tools for substring search, recent listing, full article retrieval, and theme summaries.-