wg21-wiki-mcp
A local MCP server that gives an LLM agent read access to the WG21 (ISO C++) committee wiki as a verifiable source of truth, requiring authentication and providing exact wikitext with provenance.
README
wg21-wiki-mcp
A local Model Context Protocol (stdio) server
that gives an LLM agent read access to the WG21 (ISO C++) committee wiki at
wiki.isocpp.org as a verifiable source of truth.
The committee wiki requires a login even to read, and its agendas, straw polls, documents, and subgroup pages are otherwise hard to reach from an agent. This server authenticates with your credentials, fetches pages live over the MediaWiki API, and caches them - returning the exact wiki text with a clickable URL and revision id so every answer can be verified.
Access requires WG21 membership. This tool stores nothing confidential in its source and never logs page content. See SECURITY.md.
What it does
- Verifiable content. Tools return wikitext byte-for-byte with provenance
(canonical URL, permanent
oldidURL,revid, last-edit time). Nothing is summarized or reformatted by the server. - Authentication, no browser. Prefers a MediaWiki bot password; falls back to your normal account via headless SSO (SimpleSAMLphp) if no bot password is set. The working path is pinned and reused; sessions that drop are transparently re-established.
- Shared, meeting-aware cache. A cross-process SQLite cache in
~/.isocpp.wiki/is shared by all your local agents. Pages are cached for a week normally and re-checked hourly during the three-times-a-year meetings (detected from the public meetings calendar). - Token-lean tools with opaque-cursor pagination and UTF-8-safe chunking of long pages.
Install
This package is not published to PyPI yet; install it from the GitHub source.
Pin a released version (recommended for reproducibility) with @v0.1.0, or
track the latest release with @master (the release branch). The simplest
option is uv, which can run it without a manual
install:
# pinned release (reproducible)
uvx --from git+https://github.com/cppalliance/wg21-wiki-mcp.git@v0.1.0 wg21-wiki-mcp
# or always the latest release (master tracks releases; --refresh bypasses uv's cache)
uvx --refresh --from git+https://github.com/cppalliance/wg21-wiki-mcp.git@master wg21-wiki-mcp
Or install from git with pip/pipx:
pipx install "git+https://github.com/cppalliance/wg21-wiki-mcp.git@v0.1.0"
# or, into a venv:
pip install "git+https://github.com/cppalliance/wg21-wiki-mcp.git@v0.1.0"
Or from a local clone (for development):
git clone https://github.com/cppalliance/wg21-wiki-mcp
cd wg21-wiki-mcp
pip install -e ".[dev]"
Requires Python 3.10+. Works on Windows, macOS, and Linux.
Configure
The wiki base URL is fixed (https://wiki.isocpp.org), so only credentials are
needed. With an MCP host, pass them in the server's launch env block. Since the
package is not on PyPI, run it from GitHub with uvx:
{
"mcpServers": {
"wg21-wiki": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/cppalliance/wg21-wiki-mcp.git@v0.1.0",
"wg21-wiki-mcp"
],
"env": {
"WIKI_BOT_USERNAME": "YourAccount@yourbot",
"WIKI_BOT_PASSWORD": "the-bot-password"
}
}
}
}
To always run the newest release instead of a pinned one, replace @v0.1.0
with @master and add "--refresh" as the first entry in args (so uv
re-resolves the branch rather than reusing its cache). Pinning a tag is
recommended for a source-of-truth tool so behavior is reproducible.
If you installed the console script (via pipx/pip), use "command": "wg21-wiki-mcp"
with no args instead. To use your normal account instead of a bot password,
supply WIKI_USER_USERNAME / WIKI_USER_PASSWORD (requires that MFA is not
enabled). If both are present, the bot password is used by default. See
.env.example for optional tuning (cache directory, TTLs,
meeting overrides). A bot password is created at Special:BotPasswords with the
Read grant and is the recommended, revocable option.
Tools
| Tool | Purpose |
|---|---|
search_wiki |
Full-text search; returns titles, API snippets (non-verbatim), and URLs. |
get_page |
Verbatim wikitext for a page or section, with provenance; chunked if large. |
list_pages |
Enumerate page titles in a namespace. |
list_namespaces |
List content namespaces and their numeric ids. |
list_meetings |
List discovered meetings (newest first); flags the active one. |
get_meeting_overview |
A meeting's landing page plus its subpage index. |
get_meeting_sessions |
Raw materials (agenda time slots + relevant pages) to compose a schedule. |
get_recent_changes |
Recent edits/new pages, optionally by namespace or since a time. |
wiki_status |
Auth path, meeting-aware TTL state, cache stats (no wiki content). |
Quickstart (programmatic)
from wg21_wiki_mcp.config import Config
from wg21_wiki_mcp.context import ServerContext
from wg21_wiki_mcp import tools
ctx = ServerContext.create(Config.from_env())
# Search, then read a page verbatim with a verifiable URL.
hits = tools.search_wiki(ctx, "some topic", limit=5)
page = tools.get_page(ctx, hits.hits[0].title)
print(page.provenance.url, page.provenance.revid)
print(page.content) # exact wikitext
Dependencies
mcp- the MCP/FastMCP server framework.requests+mwclient- MediaWiki API access (re-login, batched fetch).beautifulsoup4+lxml- parse the SSO login form for headless user auth.pydantic- structured tool outputs.python-dotenv- optional.envloading for local dev.filelock- cross-process single-flight page locking.
Documentation
- ARCHITECTURE.md - design, data flow, parse-vs-offload policy, what may break, future work.
- CONTRIBUTING.md - dev setup, tests, confidentiality rules, where to start reading.
- SECURITY.md - credential handling and confidentiality.
License
Boost Software License 1.0 - see LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。