Weltsuche

multilingual web search and page reading for claude code, as a plugin: an mcp server plus the skill that tells claude how to use it.

• search per language and region: google (through startpage), baidu, yandex, naver, duckduckgo, brave and wikipedia, merged and ranked.
• read pages with a browser fingerprint: articles, pdfs, legacy charsets, anubis-gated sites. walled sites (reddit, x, youtube with transcripts, instagram, tiktok, threads, bluesky, telegram, stack exchange, naver blog) are read through their apis, embeds or public frontends.
• search code, packages, developer communities and papers: github, gitee, codeberg, gitlab, grep.app, npm, crates.io, hacker news, stack overflow, juejin, csdn, v2ex, qiita, zenn, velog, habr, pubmed, arxiv.
• search shops and stays with their own filters: amazon, ebay, kleinanzeigen, airbnb.
• fetch with focus returns only the passages of a long page or pdf that mention a few words.
• no accounts, no api keys. github code search borrows the github cli's login when there is one.

claude does the translating: the question can arrive in english, and claude writes the queries in chinese, russian, korean, hebrew or japanese itself. the skill tells it when a topic is worth another language, how deep to go, and to keep using its other tools alongside.

install

you need claude code and uv. uv brings python and every dependency itself: no venv, no pip, no config to write.

1. install uv, unless uv --version already answers. these are uv's own installers.

   macos and linux:

   curl -LsSf https://astral.sh/uv/install.sh | sh
   

   windows, in powershell:

   powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
   

   then open a new terminal, so that uv is on the PATH.

2. in the folder that holds the weltsuche checkout, add it to claude code and build its environment once (all platforms):

   claude plugin marketplace add ./weltsuche
   claude plugin install weltsuche@weltsuche
   claude mcp list
   

   claude mcp list starts the server once, which builds its environment (about 15 seconds the very first time) and shows plugin:weltsuche:weltsuche ... Connected. without it, the first session may start before the environment is ready.

3. start a new claude code session (in a running one: /reload-plugins).

on first start weltsuche creates its config.py with defaults that work.

update and uninstall

• after git pull (or any edit): /reload-plugins in a session, or a new session. claude code loads this plugin in place from the checkout, so no version bump or reinstall is needed.
• uninstall: claude plugin uninstall weltsuche@weltsuche, then claude plugin marketplace remove weltsuche. uninstalling deletes the plugin's data folder (config, cache, cookies) unless you pass --keep-data.

use it from a terminal

in this folder (uv builds the environment on first use):

uv run python -m weltsuche search "topping dx5 ii review" --lang en
uv run python -m weltsuche search "拓品 DX5 II 评测" --lang zh
uv run python -m weltsuche search "thinkpad t14" --engines shopping --lang de --param price_max=400
uv run python -m weltsuche fetch https://youtu.be/dQw4w9WgXcQ --lang en
uv run python -m weltsuche status
uv run python -m weltsuche check

search takes --lang, --region, --engines a,b, --max n, --param name=value (repeatable); fetch takes --lang, --max-chars n, --offset n, --raw, --focus "words". --data DIR picks another data folder, --debug logs everything.

if something does not work

symptom	what to do
/mcp shows weltsuche as failed right after installing	the first start builds the environment; run claude mcp list once in a terminal, or reconnect it in /mcp
uv: command not found in the mcp log	open a new terminal (uv's folder joins PATH there) and restart claude code
a site comes back with error	read tried in the reply: every route and why it failed. uv run python -m weltsuche fetch URL shows the same in a terminal
engines resting	a rate limit or captcha; they come back on their own. uv run python -m weltsuche status lists them
anything else	data/weltsuche.log (or weltsuche.log in the plugin's data folder, ~/.claude/plugins/data/weltsuche-weltsuche) has every request and every traceback

configuring

config.example.py lists every setting with its default and a comment. weltsuche copies it to config.py in its data folder on first start, with every setting commented out; remove the # in front of a setting to change it. uv run python -m weltsuche check names keys weltsuche does not know; a value of the wrong shape stops the start with an error that names the key. there are no environment variables and no secrets.

more

• docs/reference.md: the tools and their failures, how pages and walled sites are read, every engine and its filters, how weltsuche stays welcome, every setting, tools that work well with it, and the repository layout.
• docs/knowledge.md: maintainer notes: why it is built this way, what every engine and site does to a plain client, the gotchas and the open issues.