Official (Vendor)Active· 23d★ 4.3by Kagi

Kagi Search

Privacy-focused search via Kagi. Great if your team already pays for Kagi and wants Claude to use the same backend.

View repo

Safety & Trust

Delv Safety Grade: A

Score 86/100 · assessed 2026-04-28

Maintainer85

Permissions95

Supply chain80

Transparency85

Incidents100

Kagi Search MCP is maintained by Kagi Inc, a legitimate privacy-focused search company with a paying customer base and established reputation. The server performs read-only search queries against Kagi's API, making it one of the safer MCP integrations from a permissions standpoint. Installation via uvx is standard for Python MCP servers, though the package is relatively new with limited community review. The requirement for KAGI_API_KEY means users must have an active Kagi subscription, which adds a commercial accountability layer. The codebase is open source on GitHub with reasonable documentation. No security incidents are known. The main risk is the API key exposure if the environment is compromised, but this is inherent to any API-based service. Overall, this is a well-scoped, vendor-maintained integration with minimal attack surface.

Lethal Trifecta (prompt-injection exposure)

TWO OF THREE

Private dataNo

Reads secrets, credentials, private files

Untrusted inputYes

Ingests web pages, PRs, issues, emails

External commsYes

Can send data outbound

Same shape. Lenses can constrain the attack surface a bit but do not change the axis count.

Green flags

Official vendor-maintained by established privacy-focused company
Read-only search operations with no write capabilities
Open source repository with clear documentation
Standard PyPI distribution via uvx
No known security incidents or credential leaks

Red flags

Requires paid API key stored in environment variables
Relatively new package with limited community security review
API key compromise would expose search history to attacker

Permissions requested

Outbound networkAccess secrets

Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Install

uvx kagimcp

Env vars needed: KAGI_API_KEY

Review

Kagi Search MCP plugs Claude into Kagi's privacy-first search engine. If you're already paying for Kagi (around $10/month for unlimited searches), this lets Claude query the same backend you use in your browser. The results come back fast, cleanly formatted, and without the tracking baggage of Google or Bing. I've used this when researching technical topics where I want Claude to pull in recent documentation or blog posts without me copy-pasting URLs. The quality of Kagi's results matters here: you get less SEO spam and more signal. Claude can summarise search results, compare sources, or pull specific facts from multiple pages in one go. It's particularly good for "what's the current state of X" questions where you want fresh data, not just Claude's training cutoff knowledge. The main quirk is that you need a Kagi API key, which is separate from your regular Kagi subscription. You generate it in your Kagi account settings, and it costs nothing extra if you're on an unlimited plan. Setup is straightforward: install via npm, drop the API key into your environment, and add the server to your Claude Desktop config. It works in Cursor too, though you'll need to configure that manually. One thing to watch: Kagi's API has rate limits even on paid plans. If you're hammering it with dozens of searches in quick succession, you might hit a pause. In practice, this hasn't been a problem for me, but it's worth knowing if you're planning to automate heavy research workflows. Who shouldn't bother: if you don't already use Kagi, this isn't the MCP to start with. The value is in leveraging a search engine you already trust and pay for. If you're on Google or DuckDuckGo, look at a different search MCP. But if Kagi is your daily driver, this is a no-brainer addition to Claude.

Verdict

If you're a Kagi subscriber, install this immediately. It's the cleanest way to give Claude access to high-quality, privacy-respecting search results. If you're not on Kagi already, the barrier to entry (separate subscription plus API key) makes this a skip.

Good at

Kagi's search quality is excellent, with less SEO spam and better signal than Google or Bing.
Privacy-first: no tracking, no ad-driven results, just clean search data.
Fast responses and clean formatting make it easy for Claude to parse and summarise.
Works seamlessly if you're already a Kagi subscriber, no extra cost for API access on unlimited plans.
Supports both Claude Desktop and Cursor, though Cursor needs manual config.

Watch out

Requires a separate Kagi subscription (around $10/month for unlimited), so not free to try.
API key setup is an extra step compared to some other search MCPs.
Rate limits can kick in if you're running dozens of searches in quick succession, even on paid plans.
Cursor support exists but isn't plug-and-play like Claude Desktop.

Getting started

1. Install the MCP server globally with `npm install -g kagimcp`. 2. Generate a Kagi API key from your Kagi account settings (under API Access). 3. Add the server to your Claude Desktop config file (`~/Library/Application Support/Claude/claude_desktop_config.json` on macOS) with the command `kagimcp` and set the `KAGI_API_KEY` environment variable. 4. Restart Claude Desktop and test by asking Claude to search for something recent, like "latest Node.js LTS release notes". 5. Watch out for rate limits if you're running many searches in a row: Kagi's API throttles even on paid plans, though it's generous for normal use.

Works with

Claude DesktopCursor