Using Linkly AI CLI

Introduction to Linkly AI CLI

Linkly AI CLI is a command-line tool that connects to Linkly AI Desktop’s MCP service, allowing you to search, browse, and read local documents from the terminal. It also serves as a bridge between AI Agents (such as Claude Desktop, Cursor) and Linkly AI.

Terminal Search

Search your documents directly from the command line — ideal for developers and power users

MCP Bridge

Run in stdio MCP mode, enabling Claude Desktop, Cursor, and other AI tools to call Linkly AI

Installation

macOS / Linux
Windows
Cargo

Run in your terminal:

curl -sSL https://updater.linkly.ai/cli/install.sh | sh

Or install via Homebrew:

brew tap LinklyAI/tap
brew install linkly

Run in PowerShell:

irm https://updater.linkly.ai/cli/install.ps1 | iex

Install from crates.io (requires Rust toolchain):

cargo install linkly-ai-cli

After installation, verify the installation:

linkly --version

By default, the CLI discovers and connects to the local Linkly AI Desktop app via ~/.linkly/port. You can also connect to a remote device via LAN or the cloud tunnel — see Connection Modes below.

Usage

The CLI follows a search → grep or outline → read progressive workflow: first search to find target documents, then use grep to find patterns or view the outline to understand the structure, and finally read the specific content. When the user describes a container (“in my Notion notes”, “in my Dropbox papers folder”) whose actual path is unknown, call find-paths before search to discover the path.

Every successful command output ends with a [meta] now=2026-05-08T...Z UTC timestamp line (or a top-level _meta.now field in JSON mode). This is metadata Desktop provides to AI assistants for computing relative dates like “last month” — human users can ignore it; for scripting, you may want to filter out the last line before further parsing.

Check Connection Status

linkly status

Returns the running status, version number, number of indexed documents, and indexing status of Linkly AI Desktop.

Search Documents

linkly search "keywords or phrases"

Searches your local documents and returns the most relevant results, including title, path, relevance score, and content snippet. Common parameters:

# Limit the number of results (default 20, max 50)
linkly search "API design" --limit 5

# Filter by document type
linkly search "meeting minutes" --type pdf,docx

# Output in JSON format (suitable for scripting)
linkly search "budget report" --json

# Limit by time window (Q3 2024 quarterly reports)
linkly search "quarterly report" --modified-after 2024-07-01 --modified-before 2024-09-30

# Sort by time ("the latest", "the earliest" — when there's no fixed window)
linkly search "weekly retro" --time-sort newest --limit 5

# Search audio and video by what was said in them
linkly search "we decided to postpone" --type audio,video

# Search your notes instead of your files
linkly search "onboarding checklist" --scope notes

--modified-after / --modified-before accept ISO 8601 UTC: a bare date 2024-01-01 (treated as 00:00:00Z) or a full RFC 3339 timestamp 2024-01-01T00:00:00Z. --time-sort accepts newest / oldest; omit it to keep the default BM25 + vector relevance order.

--scope notes restricts results to your notes and ignores --library and --path-glob — those filters are silently dropped rather than rejected.

View Document Outline

linkly outline <DOC_ID>

Retrieves the structured outline and metadata of a document. DOC_ID is obtained from search results. You can view multiple documents at once, or pipe the IDs in with -:

linkly outline id1 id2 id3

# Outline everything a search found, without a shell loop
linkly search "architecture" --json | jq -r '.results[].doc_id' | linkly outline -

The outline feature works best with Markdown, DOCX, PowerPoint (PPTX), and EPUB documents, as their heading structures can be parsed. For plain text or PDFs without bookmarks, it’s recommended to use the read command directly.

Search Patterns in Documents

linkly grep <PATTERN> <DOC_ID>...

Search for regex pattern matches inside one or more documents. Use when you need to find specific text (terms, names, dates, identifiers, etc.):

# Find a pattern in a document
linkly grep "useState" 456

# Case-insensitive with context lines
linkly grep "error|warning" 1044 -C 3 -i

# Count matches only
linkly grep "TODO" 591 --mode count

# Paginate matches (skip first 20, show next 20)
linkly grep "import" 302 --offset 20 --limit 20

# Grep across everything a search found
linkly search "deployment" --json | jq -r '.results[].doc_id' \
  | linkly grep "docker|kubernetes" -

Read Document Content

linkly read <DOC_ID>...

Reads the full content of one or more documents, outputting text with line numbers. For long documents, you can read in pages:

# Start from line 50, read 100 lines
linkly read <DOC_ID> --offset 50 --limit 100

# Include the OCR text of images referenced in the page
linkly read <DOC_ID> --image-text full

# Read several documents at once, or pipe the IDs in
linkly read id1 id2 id3
linkly search "onboarding" --json | jq -r '.results[].doc_id' | linkly read -

With --json, several documents print as JSON Lines — one object per line. A single ID still prints a single object, so existing scripts are unaffected.

Pagination strategy: By default, 200 lines are read per request (max 500 lines). For long documents, adjust --offset to read progressively:

linkly read <DOC_ID> --offset 1 --limit 200    # Lines 1-200
linkly read <DOC_ID> --offset 201 --limit 200  # Lines 201-400
linkly read <DOC_ID> --offset 401 --limit 200  # Lines 401-600

Find Paths (find-paths)

linkly find-paths --patterns <keyword,keyword,...>

Fuzzy-matches keywords against the file path field of indexed documents, aggregates matches at folder granularity, and returns the top folder candidates. It is positioned as a helper for search: when the user names a container (“in my Notion notes”, “in my Dropbox papers folder”) but you don’t know its on-disk path, call find-paths first, then pass a distinctive segment of the returned path to search as --path-glob. When a folder name contains glob metacharacters (* ? [), use the returned path_glob field directly — it is already escaped to match that folder literally. Two-step workflow:

# Step 1: locate the real path
linkly find-paths --patterns Notion,notion --limit 5
# Suppose it returns .../Documents/Notion-Export-abc/workspace (1240 files)

# Step 2: search within that container
linkly search "shopping receipt" --path-glob "*Notion-Export*"

Variant matching: --patterns takes a comma-separated list of keywords, OR-matched against the path. Pass several variants in one call (translation pairs, casing, real app/SDK identifiers when known) to maximise first-pass recall:

linkly find-paths --patterns Notion,notion,Notion-Export
linkly find-paths --patterns Slack,slack --library work-notes

find-paths is a “find folders” tool, not a “find files” tool: only matches on directory segments count. If keywords match only the filename segment (an “orphan file”), they are silently dropped. If a query returns zero folders even though you expect matching files, fall back to calling linkly search directly without --path-glob.

Notes

Linkly AI keeps short Markdown notes in your library folder. They are ordinary local files — never uploaded — and they are indexed like any other document.

# List them, newest first
linkly list --scope notes

# Filter by tag
linkly list --scope notes --tags project,urgent

# Write one
linkly note-save --mode create --content "Remember to renew the certificate" --tags ops

# Pipe a long body in instead of quoting it
some-command | linkly note-save --mode create --content -

Editing an existing note needs its note_id and current version, both of which linkly list --scope notes returns:

linkly note-save --mode edit \
  --note-id <uuid> --base-version <version> \
  --tags ops --content "Updated text"

Inline #tags in the note body are the note’s tags — --tags only adds (remove a tag by deleting its #tag marker from the content). Desktops older than 0.11.0 instead require --tags on edit and treat it as the full replacement set. --base-version is a concurrency check: if the note changed since you read it, the command fails with NOTE_VERSION_CONFLICT instead of overwriting. Note content is a restricted Markdown subset (paragraphs, bold, strikethrough, lists); headings, code, links and tables are rejected.

Shell Completions

linkly completions <SHELL>

Prints a completion script for bash, zsh, fish, powershell or elvish.

zsh
bash
fish
PowerShell

mkdir -p ~/.zfunc
linkly completions zsh > ~/.zfunc/_linkly
# Add to ~/.zshrc if not already there:
#   fpath=(~/.zfunc $fpath)
#   autoload -Uz compinit && compinit

linkly completions bash > /usr/local/etc/bash_completion.d/linkly

linkly completions fish > ~/.config/fish/completions/linkly.fish

linkly completions powershell | Out-String | Invoke-Expression

Open a new shell afterwards. The script is static — it never contacts Linkly AI Desktop, so it works with the app closed and can’t slow down your prompt.

MCP Mode

linkly mcp

Runs as a stdio MCP server, exposing Linkly AI’s tools to MCP-compatible AI clients. The upstream decides what the client can reach:

Command	Reaches
`linkly mcp`	The local desktop app — your local content only
`linkly mcp --endpoint <url>`	A desktop app on the LAN — same content boundary
`linkly mcp --remote`	The cloud gateway — local content and linked cloud libraries

--remote is the only bridge mode that can reach cloud libraries. It needs an API key saved first (see Remote Mode).

Configure Claude Desktop and other local AI apps: Add the following to the configuration file of Claude Desktop or similar apps:

macOS / Linux
Windows

Edit ~/.config/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "linkly-ai": {
      "command": "linkly",
      "args": ["mcp"]
    }
  }
}

Edit %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "linkly-ai": {
      "command": "linkly",
      "args": ["mcp"]
    }
  }
}

Configure Cursor: In Cursor, open Settings → MCP Servers → Add Server, and add:

Name: linkly-ai
Command: linkly mcp

Update CLI

linkly self-update

Automatically checks for and updates to the latest version. The CLI also checks for updates in the background on each launch and will prompt you to run this command if a new version is available.

Connection Modes

The CLI supports three ways to connect to your Linkly AI knowledge base:

Mode	Flags	How it works
Local	(default, no flags needed)	Auto-discovers the desktop app via `~/.linkly/port`
LAN	`--endpoint <url> --token <token>`	Direct connection to another device on the local network
Remote	`--remote`	Connects via `https://mcp.linkly.ai` cloud tunnel using API Key

Local Mode (default)

No extra flags needed. The CLI reads ~/.linkly/port to find the running desktop app:

linkly search "machine learning"
linkly status

LAN Mode

Connect to a Linkly AI instance running on another device in your local network. The token can be found in the desktop app under Settings → MCP:

linkly search "report" --endpoint http://192.168.1.100:60606/mcp --token your_lan_token
linkly status --endpoint http://192.168.1.100:60606/mcp --token your_lan_token

Remote Mode

Connect to your knowledge base from anywhere via the cloud tunnel. First, save your API key (from linkly.ai/dashboard):

linkly auth set-key lkai_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Then use --remote with any command:

linkly search "machine learning" --remote
linkly status --remote

Check or clear the stored key at any time:

linkly auth status   # which key is in use, whether it's valid, and your plan
linkly auth logout   # remove the stored credentials

--endpoint and --token are required together for LAN access. They cannot be combined with --remote. For remote access, use linkly auth set-key to save your API key.

Parameter Reference

Global Options

Option	Scope	Description
`--endpoint <URL>`	LAN	Connect to a specific MCP endpoint (e.g. `http://192.168.1.100:60606/mcp`), requires `--token`
`--token <token>`	LAN	Bearer token for LAN authentication (required with `--endpoint`, conflicts with `--remote`)
`--remote`	Remote	Connect via cloud tunnel at `https://mcp.linkly.ai` (conflicts with `--endpoint`)
`--json`	All commands	Output in JSON format (suitable for scripting and automation)
`--exit-code`	All commands	Distinguish “found nothing” (exit 1) from “failed” (exit 2) — see Exit Codes
`-V, --version`	—	Display CLI version number
`-h, --help`	—	Display help information

--endpoint, --token and --remote are available on the document commands (search, grep, outline, read, list, note-save, find-paths, explore, list-libraries) plus status and doctor. The mcp command accepts --endpoint or --remote, but not --token. --json and --exit-code are available everywhere.

Exit Codes

By default the CLI uses the conventional two values: 0 on success, 1 on failure. Note that “success” includes finding nothing — a search with no hits still exits 0. Pass --exit-code to tell the two apart:

Code	Meaning
`0`	Ran and produced at least one result
`1`	Ran fine but matched nothing
`2`	Failed (connection, authentication, bad arguments, errors)

# Only opens the report when the search actually found something
linkly search "quarterly report" --exit-code && open-report

The flag is opt-in because it changes what 1 means. Without it, 1 is “failed” — which is what existing scripts test for.

search Parameters

Parameter	Description	Default
`<QUERY>`	Search keywords or phrase (required)	—
`--limit <N>`	Maximum number of results (1-50)	20
`--type <TYPES>`	Filter by document type (comma-separated): `pdf`, `docx`, `pptx`, `epub`, `md`, `txt`, `html`, `image`, `audio`, `video`	All
`--library <NAME>`	Restrict search to a specific library	—
`--path-glob <PATTERN>`	Filter by file path; substring-matched (may appear anywhere, no leading/trailing `*` needed). When the actual path is unknown, run `linkly find-paths` first	—
`--modified-after <ISO>`	Inclusive lower bound on modification time. ISO 8601 UTC: bare date `2024-01-01` or full `2024-01-01T00:00:00Z`	—
`--modified-before <ISO>`	Inclusive upper bound on modification time. Same format as `--modified-after`	—
`--time-sort <MODE>`	Time-based reordering: `newest` / `oldest`. Omit to keep relevance ordering	—
`--scope <SCOPE>`	`folder` searches all indexed content; `notes` restricts results to your notes and ignores `--library` / `--path-glob`	folder
`--tags <LIST>`	Comma-separated note tags; returns only documents carrying all of them	—

find-paths Parameters

Parameter	Description	Default
`--patterns <LIST>`	Comma-separated keyword list (required). Multiple keywords are OR-matched; pass several variants in one call (translation pairs, casing). Case-insensitive for ASCII; CJK is literal.	—
`--library <NAME>`	Restrict to a specific library	—
`--limit <N>`	Maximum number of candidate folders (1-50)	10

outline Parameters

Parameter	Description	Default
`<ID>...`	Document IDs (required, supports multiple). Pass `-` to read the IDs from stdin, one per line	—

grep Parameters

Parameter	Description	Default
`<PATTERN>`	Regular expression pattern (required)	—
`<DOC_ID>...`	Document IDs to search (required, supports multiple). Pass `-` to read the IDs from stdin	—
`-C, --context`	Lines of context before and after	3
`-B, --before`	Lines of context before each match	—
`-A, --after`	Lines of context after each match	—
`-i`	Case-insensitive matching	—
`--mode`	Output mode: `content` or `count`	content
`--limit`	Maximum matches (max 100)	20
`--offset`	Number of matches to skip	0
`--fuzzy-whitespace`	Fuzzy whitespace matching: `true` to force on, `false` to force off, omit for auto (PDF on, others off)	auto

read Parameters

Parameter	Description	Default
`<ID>...`	Document IDs (required, supports multiple). Pass `-` to read the IDs from stdin	—
`--offset <N>`	Starting line number (from 1)	1
`--limit <N>`	Number of lines to read (max 500)	200
`--image-text <DETAIL>`	Detail for images referenced in the page: `none` (mapping only), `abstract` (excerpt), `full` (inline OCR text)	abstract

list Parameters

Parameter	Description	Default
`--scope <SCOPE>`	Container to list: `folder` (indexed files under a directory), `library` (one library’s files), or `notes` (required)	—
`--library <LIB>`	Which library, for `--scope library` — a name, `local://<id>`, or `cloud://<owner>/<slug>` (remote)	—
`--path <DIR>`	Directory to list — an absolute path locally, a relative prefix for cloud libraries	—
`--type <LIST>`	Comma-separated document types (`folder` / `library` only), e.g. `pdf,docx`	all
`--modified-after <DATE>`	Only items modified on or after this date (`folder` / `library` only). ISO 8601, e.g. `2026-01-01`	—
`--modified-before <DATE>`	Only items modified on or before this date. Same format	—
`--tags <LIST>`	Comma-separated tags (`notes` only); returns only items carrying all of them	—
`--limit <N>`	Maximum items (max 200; capped at 50 while snippets are on)	50
`--offset <N>`	Pagination offset	0
`--sort <ORDER>`	`recent` (newest first), `oldest`, or `name`	recent
`--snippet` / `--no-snippet`	Force per-item snippets on / off (mutually exclusive). Snippets default on for `notes`, off for `folder` / `library`	per scope

note-save Parameters

Parameter	Description	Default
`--mode <MODE>`	`create` writes a new note; `edit` rewrites one (requires `--note-id` and `--base-version`)	—
`--content <TEXT>`	Note body (required). Pass `-` to read it from stdin	—
`--note-id <UUID>`	Which note to edit, from `linkly list --scope notes`	—
`--base-version <HASH>`	The note’s current version, from `linkly list --scope notes`. Rejects the write if the note changed meanwhile	—
`--tags <LIST>`	Comma-separated tags to add — remove a tag by deleting its `#tag` marker from the content	—
`--app-name <NAME>`	Display name of the application hosting the conversation, shown as the note’s source badge. Max 64 characters	—

completions Parameters

Parameter	Description	Default
`<SHELL>`	`bash`, `zsh`, `fish`, `powershell` or `elvish` (required)	—

Getting Started

Core Features

Connect Your AI Tools

For AI Agents

Using Linkly AI CLI

Introduction to Linkly AI CLI

Terminal Search

MCP Bridge

Installation

Usage

Check Connection Status

Search Documents

View Document Outline

Search Patterns in Documents

Read Document Content

Find Paths (find-paths)

Notes

Shell Completions

MCP Mode

Update CLI

Connection Modes

Local Mode (default)

LAN Mode

Remote Mode

Parameter Reference

Global Options

Exit Codes

search Parameters

find-paths Parameters

outline Parameters

grep Parameters

read Parameters

list Parameters

note-save Parameters

completions Parameters

​Introduction to Linkly AI CLI

Terminal Search

MCP Bridge

​Installation

​Usage

​Check Connection Status

​Search Documents

​View Document Outline

​Search Patterns in Documents

​Read Document Content

​Find Paths (find-paths)

​Notes

​Shell Completions

​MCP Mode

​Update CLI

​Connection Modes

​Local Mode (default)

​LAN Mode

​Remote Mode

​Parameter Reference

​Global Options

​Exit Codes

​search Parameters

​find-paths Parameters

​outline Parameters

​grep Parameters

​read Parameters

​list Parameters

​note-save Parameters

​completions Parameters

Introduction to Linkly AI CLI

Installation

Usage

Check Connection Status

Search Documents

View Document Outline

Search Patterns in Documents

Read Document Content

Find Paths (find-paths)

Notes

Shell Completions

MCP Mode

Update CLI

Connection Modes

Local Mode (default)

LAN Mode

Remote Mode

Parameter Reference

Global Options

Exit Codes

search Parameters

find-paths Parameters

outline Parameters

grep Parameters

read Parameters

list Parameters

note-save Parameters

completions Parameters