Linkly AI CLI

Command-line interface for Linkly AI — search your local documents from the terminal.

The CLI connects to the Linkly AI desktop app's MCP server, giving you fast access to your indexed documents without leaving the terminal.

Prerequisites

By default, the Linkly AI desktop app must be running with MCP server enabled. The CLI automatically discovers the app via ~/.linkly/port. Alternatively, use LAN mode (--endpoint + --token) or Remote mode (--remote with a saved API key) — see Connection Modes.

Installation

macOS / Linux

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

Windows (PowerShell)

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

Homebrew (macOS / Linux)

brew tap LinklyAI/tap
brew install linkly

Cargo

cargo install linkly-ai-cli

GitHub Releases

Pre-built binaries for all platforms are available on the Releases page.

Platform	File
macOS (Apple Silicon)	`linkly-aarch64-apple-darwin.tar.gz`
macOS (Intel)	`linkly-x86_64-apple-darwin.tar.gz`
Linux (x86_64)	`linkly-x86_64-unknown-linux-gnu.tar.gz`
Linux (ARM64)	`linkly-aarch64-unknown-linux-gnu.tar.gz`
Windows (x64)	`linkly-x86_64-pc-windows-msvc.zip`

From Source

cargo install --path .

Usage

Find Paths

Locate folder candidates when you only know a fuzzy or cross-language container name (e.g. "WeChat" but the actual path contains xinWeChat). Pass several variants in --patterns so the keywords are OR-matched in a single call. Output is intended as a --path-glob source for linkly search.

linkly find-paths --patterns "WeChat,微信,wxid"
linkly find-paths --patterns "Notion,notion" --library my-research
linkly find-paths --patterns "Dropbox" --limit 5

Option	Description
`--patterns <list>`	Comma-separated keywords (OR-matched); ASCII case-insensitive, CJK literal
`--library <name>`	Restrict to a specific library by name
`--limit <N>`	Maximum folder candidates returned (default: 10, max: 50)

Search Documents

linkly search "machine learning"
linkly search "API design" --limit 5
linkly search "notes" --type pdf,md,docx,pptx,epub
linkly search "attention" --library my-research
linkly search "transformer" --path-glob "*.pdf"
linkly search "quarterly report" --modified-after 2024-01-01 --modified-before 2024-12-31
linkly search "weekly notes" --time-sort newest --limit 10

Option	Description
`--limit <N>`	Maximum results (default: 20, max: 50)
`--type <types>`	Filter by document types, comma-separated (e.g. `pdf,md,docx,pptx,epub,txt,html`)
`--library <name>`	Restrict search to a specific library by name
`--path-glob <pat>`	Glob substring-matched against the file path — may appear anywhere, no leading/trailing `` needed. `` matches any chars (incl. `/`), `?` one char. Examples: `*.pdf`, `papers`, `/Users/me/notes/` (a full directory path scopes to that dir). When the actual path is unknown, run `linkly find-paths` first.
`--modified-after <iso>`	Inclusive lower bound on file modification time. Accepts a bare date (`2024-01-01`) or RFC 3339 (`2024-01-01T00:00:00Z`). UTC. Use for explicit windows like "after January 2024".
`--modified-before <iso>`	Inclusive upper bound. Same format as `--modified-after`.
`--time-sort <mode>`	Reorder by modification time: `newest` or `oldest`. Omit (default) to keep BM25 + vector relevance ordering. Use `newest` for "recent / latest" intent without a fixed window.

View Document Outline

Get structural outlines for one or more documents (IDs come from search results):

linkly outline <doc-id>
linkly outline <id1> <id2> <id3>

Locate Lines in a Document

linkly grep "pattern" <doc-id>
linkly grep "error|warning" <doc-id> -C 3 -i
linkly grep "TODO" <doc-id> --mode count

Option	Description
`-C, --context`	Lines of context before and after each match
`-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`
`--limit`	Maximum matches (default: 20, max: 100)

Read Document Content

linkly read <doc-id>
linkly read <doc-id> --offset 50 --limit 100

Option	Description
`--offset <N>`	Starting line number (1-based)
`--limit <N>`	Number of lines to read (max: 500)

List Container Contents

Enumerate a container without full-text matching — indexed files under a directory, one library's files, or your notes:

linkly list --scope folder --path /Users/me/docs
linkly list --scope folder                          # all watched roots
linkly list --scope library --library "My Library"
linkly list --scope notes --tags project --json

Option	Description
`--scope <s>`	Required: `folder`, `library`, or `notes`
`--library <ref>`	Which library to list (`--scope library`; a name or `local://<id>`)
`--path <dir>`	Absolute directory to list (`--scope folder`, or inside a local library)
`--type <list>`	Filter by document types, comma-separated (folder/library)
`--modified-after/-before <t>`	Modification-time bounds, ISO 8601 UTC (folder/library)
`--tags <list>`	Filter notes by tags, comma-separated, AND semantics (notes)
`--sort <s>`	`recent` (default), `oldest`, or `name`
`--limit` / `--offset`	Pagination (default 50, max 200; capped at 50 while snippets are on)
`--snippet` / `--no-snippet`	Force per-item snippets on/off (default: notes on, folder/library off)

Notes live on the Desktop machine — with --remote the listing reaches them through the tunnel; there is no cloud notes store.

Save a Note

Create or edit a markdown card note in your Desktop's Notes folder:

linkly note-save --mode create --content "Remember this #idea"
echo "Piped body" | linkly note-save --mode create --content -
linkly note-save --mode edit --note-id <uuid> --base-version <hash> --content "New body #idea"

Option	Description
`--mode <m>`	`create` or `edit` (edit requires `--note-id` and `--base-version` together)
`--content <text>`	Markdown body without YAML front matter; `-` reads it from stdin
`--note-id <uuid>`	Which note to edit (from `linkly list --scope notes`)
`--base-version <hash>`	The version you read (compare-and-swap; from the same `list`)
`--tags <list>`	Note tags (see the caution below before using this on edit)
`--app-name <name>`	Hosting application's display name, shown as the note's source badge

Edits are compare-and-swap: a stale --base-version is rejected — re-read, merge, retry. Tags live in the note body as #tokens (the source of truth): --tags only adds tags, and you remove one by deleting its #token from the content. Caution: Desktops older than 0.11.0 instead require --tags on edit and treat it as the full replacement set (tags you omit are deleted). Requires Desktop >= 0.11.0.

Check Status

linkly status

MCP Bridge Mode

Run as a stdio MCP server for Claude Desktop, Cursor, or other MCP clients:

linkly mcp

Claude Desktop configuration (claude_desktop_config.json):

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

The bridge advertises the desktop's nine tools. With --remote ("args": ["mcp", "--remote"]) it also advertises the two cloud-only tools the gateway implements — library_search (find cloud libraries in the catalog, including ones not linked yet) and library_link (link one so search / explore / list can use it, or with action: "unlink" release a link the user named, e.g. when the Slot quota is full) — and the server instructions describe that reach. Local and LAN bridges never advertise them: the desktop does not implement them, so the tool list matches what the upstream can actually serve.

Diagnose Connection Issues

linkly doctor
linkly doctor --remote
linkly doctor --endpoint http://192.168.1.100:60606/mcp --token abc123

Runs a series of checks (port file, server reachability, auth, MCP round-trip) and reports pass/fail with actionable advice.

List Libraries

List all available knowledge libraries (useful with search --library):

linkly list-libraries
linkly list-libraries --remote

Self-Update

linkly self-update

Agent Skill

The linkly-ai skill teaches agents how to drive these tools. The CLI reports its version, installs it, and upgrades it in place:

linkly skills status    # installed version, latest published, and where it lives
linkly skills install   # download and install
linkly skills update    # upgrade an installed copy, or install a missing one

linkly status reports the skill in its main table, checked live on every run. That is the reliable place to look; everything below is the nudge for someone who did not think to look.

When the skill is missing or out of date, output carries a one-line notice addressed to the assistant reading it — what is wrong, what it costs, and to ask the user before running the fix. It leads the output rather than trailing it, so a long result cannot bury it and a client that truncates cannot drop it. Under --json it arrives as a skill_notice field; over linkly mcp it is a separate content block ahead of the answer, sent once per process.

A missing skill is reported on every run — deciding it costs one filesystem check, and throttling it means one unrelated process can silence it for every session in the next four hours. The out-of-date notices do read the update server, so those are throttled to once every four hours. To turn it all off:

export LINKLY_NO_SKILLS_HINT=1

linkly mcp inherits the environment, so setting it in an MCP client's config silences the bridge too.

update follows the shape of each install: a symlink is left to the store it points at, a git checkout is reported rather than overwritten, and a real directory is replaced with the previous copy moved aside until the new one verifies. Only the linkly-ai subdirectory is ever touched — the surrounding skills directory belongs to whatever else you have installed there.

Detection also looks at linkly-ai-skills, the directory name produced by installs made during a 37-hour window in March 2026 when SKILL.md carried the repository name in its name: field. Those copies work, so reporting them as missing would be a false alarm; status labels the path as legacy, and install / update still write only to linkly-ai.

Connection Modes

The CLI supports three connection modes:

Mode	Flags	Auth	How it works
Local	(default)	None (localhost)	Reads `~/.linkly/port`, connects to `127.0.0.1`
LAN	`--endpoint <url> --token <token>`	Bearer token from desktop app	Direct connection to a LAN device
Remote	`--remote`	API Key via `auth set-key`	Connects via `https://mcp.linkly.ai` tunnel

Note: --endpoint and --token are required together for LAN access and conflict with --remote. For remote access, use linkly auth set-key. The mcp command also accepts --endpoint alone (without --token).

Remote mode setup

# Save your API Key (from https://linkly.ai/dashboard)
linkly auth set-key lkai_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

# Search via remote tunnel
linkly search "machine learning" --remote

LAN mode with token

# Connect to another device on the same network (token from desktop Settings → MCP)
linkly search "report" --endpoint http://192.168.1.100:60606/mcp --token your_lan_token

Options

Connection options (--endpoint, --token, --remote) are available on search, find-paths, explore, grep, outline, read, list, note-save, status, doctor, and list-libraries commands. --endpoint alone is also available on mcp. --json is available on all commands.

Flag	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 `https://mcp.linkly.ai` tunnel (conflicts with `--endpoint`, requires `auth set-key`)
`--json`	Global	Output in JSON format (useful for scripting)
`-V, --version`	Global	Print version
`-h, --help`	Global	Print help

Examples

# Local search (default, requires desktop app running)
linkly search "budget report"

# Search across LAN with token
linkly search "budget report" --endpoint http://192.168.1.100:60606/mcp --token abc123

# Search via remote tunnel
linkly search "TODO" --remote

# JSON output for scripting
linkly search "TODO" --json | jq '.content'

# Pipe document content
linkly read abc123 --limit 50 | head -20

Community

Documentation — guides, integrations, and troubleshooting
Community — every official channel in one place
GitHub Issues — bugs and feature requests
YouTube · X

License

Apache-2.0

Name		Name	Last commit message	Last commit date
.github		.github
scripts		scripts
src		src
tests		tests
.gitignore		.gitignore
Cargo.lock		Cargo.lock
Cargo.toml		Cargo.toml
LICENSE		LICENSE
README.md		README.md

Navigation Menu

Uh oh!

Repository files navigation

Linkly AI CLI

Prerequisites

Installation

macOS / Linux

Windows (PowerShell)

Homebrew (macOS / Linux)

Cargo

GitHub Releases

From Source

Usage

Find Paths

Search Documents

View Document Outline

Locate Lines in a Document

Read Document Content

List Container Contents

Save a Note

Check Status

MCP Bridge Mode

Diagnose Connection Issues

List Libraries

Self-Update

Agent Skill

Connection Modes

Remote mode setup

LAN mode with token

Options

Examples

Community

License

About

Releases

Packages

Contributors

Languages

Uh oh!

Latest commit

History

Folders and files

Repository files navigation

Linkly AI CLI

Prerequisites

Installation

macOS / Linux

Windows (PowerShell)

Homebrew (macOS / Linux)

Cargo

GitHub Releases

From Source

Usage

Find Paths

Search Documents

View Document Outline

Locate Lines in a Document

Read Document Content

List Container Contents

Save a Note

Check Status

MCP Bridge Mode

Diagnose Connection Issues

List Libraries

Self-Update

Agent Skill

Connection Modes

Remote mode setup

LAN mode with token

Options

Examples

Community

License

About

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages