kagi_extract
Extract the content of a web page as markdown using the Kagi Extract API. Use this to read the full content of a page when needed.
LABSCO SUMMARY
"Open the third result" resolves without searching again.
Two tools is the entire surface and the split between them is the point: results come back numbered so a reference to one of them survives into the next turn, and the chosen URL is then fetched as markdown rather than being reasoned about from a snippet. Filters that normally live in search-operator syntax — include_domains, exclude_domains, file_type, before and after — are ordinary parameters here, so a model can narrow its own query instead of rephrasing it.
What it is
Two tools over the Kagi Search and Kagi Extract APIs: a numbered result set, and the full content of a page as markdown.
What you get
- Web results from the Kagi Search API, numbered so a later message can refer to a result by its number
- Query scoping by time — after, before, or a relative window — and by included or excluded domains, file type and lens
- A result limit and an extract count set on the same call
- Any URL fetched and converted to markdown through the Kagi Extract API
Requirements
KAGI_API_KEY for a Kagi account, and uv with a Python available (UV_PYTHON).
Setup effort
One command plus a key — uvx kagimcp, then supply credentials
USE CASES
01
Researching a question where the snippet never settles it
Search, then pull the page that looked right back as markdown and read it properly. Results are numbered, so the follow-up refers to a result rather than repeating the query.
02
Holding a search to sources you actually trust
Domain includes and excludes, file type, a lens, and an absolute or relative date window are parameters on the call, so the constraint is set by the model rather than typed into a query string.
03
Reading a page someone handed you as a link
No search step: give the URL and get markdown back, which is what you want when the link came from a colleague and the question is what it says.
2 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
kagi_search_fetch
Fetch web results for a query using the Kagi Search API. Use for general search and when the user explicitly tells you to 'fetch' results/information. Results are numbered so that a user may refer to a result by a specific number.
BEFORE YOU INSTALL
What to know first
Tools you get
2 tools when we connected. kagi_fastgpt and kagi_summarizer were removed and are planned to return, so listings showing four tools are out of date. KAGI_HIDDEN_PARAMS hides chosen search parameters from the model.
Requires
- uvx on PATH
- KAGI_API_KEY in your environment
CONTEXT COST
Light Standing cost
Rated by the size of the 2 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
Connect your client
Server setup
uvx
TERMINAL
uvx kagimcpHosted endpoint
ENDPOINT URL
https://mcp.kagi.com/mcpConnect a client to this instead of installing anything. Send your Kagi API key, from kagi.com/api/keys, as an Authorization: Bearer header. Taken from this project's own documentation.
MCP Clients
Claude Code — Remote
Run this in bash, not zsh (the macOS default): in zsh the key prompt fails and the server is saved with an empty key. The $(...) part can also be your key itself.
CLAUDE CODE — REMOTE · TERMINAL
claude mcp add kagi https://mcp.kagi.com/mcp --transport http --header "Authorization: Bearer $(read -sp 'API key: ' k; echo $k)" --scope userCodex
CODEX · TERMINAL
codex mcp add kagi --env KAGI_API_KEY=<YOUR_KAGI_API_KEY> -- uvx kagimcpClaude Desktop, Kiro
JSON
{
"mcpServers": {
"kagi": {
"command": "uvx",
"args": [
"kagimcp"
],
"env": {
"KAGI_API_KEY": "<YOUR_KAGI_API_KEY>"
}
}
}
}Claude Code — Local
CLAUDE CODE — LOCAL · TERMINAL
claude mcp add kagi -e KAGI_API_KEY="<YOUR_KAGI_API_KEY>" -- uvx kagimcpOpenCode
OPENCODE · JSON
{
"mcp": {
"kagi": {
"type": "local",
"command": [
"uvx",
"kagimcp"
],
"enabled": true,
"environment": {
"KAGI_API_KEY": "<YOUR_KAGI_API_KEY>"
}
}
}
}
