Skip to main content
Linkly AI hands its retrieval capability to AI assistants through MCP (Model Context Protocol). The setup pages for twenty-plus clients all look different, but underneath there are only four connection paths. Figure out which one you need first, then go to the page for your client — it will save you a lot of trial and error.

​
The four paths

When in doubt, pick the first one. If your client is installed on your own computer — Claude Code, Cursor, Cherry Studio and the like — local MCP is the easiest route: the MCP server is ready as soon as the desktop app starts, no account required and no internet connection needed.

​
How to decide quickly

1

Where does your client run?

On your computer (Claude Code, Cursor, Cherry Studio, Raycast…) → use local MCP; if the client only supports stdio, use the CLI bridge instead.On someone else’s server (ChatGPT, Claude.ai and other web apps) → they can’t reach your 127.0.0.1, so only the two cloud paths are available.
2

How much should the AI be able to read?

Only a specific set of documents → use a Cloud Library and push that content up. Available on free accounts, and it works even when your computer is off.Every document on your computer → use the remote access tunnel. This is a Pro feature, and it requires the desktop app to stay running with the tunnel connected.
The last two paths share the same endpoint, https://mcp.linkly.ai/mcp. They are told apart by whether a cloud library is specified at query time: specify one and the cloud copy is searched; omit it and the request routes back to your own machine by default (that is, through the tunnel). So give them different server names in your client (linkly-ai-cloud vs linkly-ai) and you can keep both configured side by side without interference.

​
Common prerequisites

Whichever path you take, these two steps are worth doing first:

Install and run the desktop app

The MCP server starts automatically with the desktop app on port 60606 by default. If that port is taken it moves to the next one; the actual port is shown under Settings → MCP.

Install Linkly AI Skills

Skills teach your AI assistant when to search, when to look at an outline, and how to page through a long document. With the same tools, an assistant that has Skills installed is noticeably better at using them.

Let AI set it up for you

Already have an AI assistant that can run commands? Send it this message and it will read the docs and walk you through the whole thing:

​
Find your client’s setup page

​
AI assistants

Claude

Web app, cloud path

Claude Cowork

Desktop assistant

ChatGPT

Web app, cloud path

OpenClaw

Open-source agent platform

Cherry Studio

Multi-model client

Raycast AI

macOS productivity tool, needs the CLI bridge

QwenPaw

Qwen agent

WorkBuddy

AI work assistant

QoderWork

Desktop AI work assistant

​
Coding agents

Cursor

AI editor

Windsurf

AI IDE

GitHub Copilot

Inside VS Code

Codex

OpenAI command-line assistant

Claude Code

Anthropic command-line assistant

Gemini CLI

Google terminal agent

ZCode

Coding agent built on GLM

Kimi Code

Kimi command-line agent

​
AI workflows and others

n8n

Workflow automation

Dify

AI app development platform

LM Studio

Local LLM client

​
Your client isn’t listed?

MCP is an open protocol — any client that supports it can connect to Linkly AI. The pages above just spell out the exact clicks for common clients; the configuration itself is universal:

Supports HTTP / StreamableHTTP

Set the URL to http://127.0.0.1:60606/mcp and the server name to linkly-ai. No auth header required.

Only supports stdio

Set the command to linkly and the argument to mcp. You’ll need to install the CLI first.
Once configured, one search is enough to verify it:

Search my documents for content about XX

If you’d like a dedicated setup page for a particular client, let us know in the community.

​
Further reading