Align CLI

@aligndottech/cli is free, open source (MIT) and needs no account. It imports decision history from the tools you already use into a local graph, and then serves that graph to any AI agent over MCP.

It is the fastest way to answer the question this product exists for: what would my agent say if it knew what we had already decided?

Install

npm install -g @aligndottech/cli
align --version

Requires Node.js 20 or later.

Ninety seconds to a graph your agent can read

align import git          # your own repo history - no token, no account
align mcp --setup         # detect your editors and write their MCP config

Then ask your agent something only the graph knows. align setup walks the same path interactively if you would rather be prompted.

Commands

Working with the graph

Command	What it does
align search "<query>"	Semantic search. --limit (10), --space <slug>
align ask "<question>"	Natural-language question over the graph. --limit (8)
align decisions list	List decisions. --platform, --status, --space, --limit (20)
align decisions show <id>	Full detail for one decision
align links list	Relationships between decisions. --relation <type>, --decision <id>
align spaces list	Spaces in the current workspace
align drift	Superseded and conflicting decisions
align export [topic]	Export decisions. --format brief|json (brief), --limit (50)
align status	Workspace and connection summary

Capturing and checking

Command	What it does
align capture <url>	Capture a decision from a URL (a Slack thread, a PR, an issue)
align check	Check your working changes against the graph. See below

Importing history

align import with no arguments runs interactively. Every source takes --approve to skip review, and --limit to bound the fetch.

Command	Extra flags	Default limit
align import git	--from, --to, --branch	500
align import github		100
align import gitlab	--domain	100
align import jira	--email, --token, --domain	100
align import confluence	--email, --token, --domain	50
align import linear		100
align import slack	--days-back (90)	50
align import teams		50
align import zoom		30
align import notion		50

Plus align import list, align import suggestions and align import scan-runs to inspect what a scan produced before approving it.

Slack is experimental and needs a user token (xoxp-), not a bot token. Create one at api.slack.com/apps with channels:read, channels:history, groups:read and groups:history.

Environment and auth

Command	What it does
align env set <prod|preview|local>	Set the default environment
align env get	Show the current environment
align login	Connect to a hosted workspace. --token <token> for non-interactive
align logout	Clear stored credentials
align whoami	Show the authenticated identity
align setup	Guided first-run setup. --local, --approve, --reset
align local start|status|reset	Manage the local embedded graph
align mcp	Run as an MCP server. --setup writes your editor config

Every command accepts --env <prod|preview|local> to override the default for one invocation.

align login is optional. Without it the CLI works entirely locally; with it, it reads and writes your organisation's shared graph.

align check - the pre-commit and CI gate

align check                                        # working changes vs the graph
align check --ci --base origin/main                # in CI, over the whole branch
align check --hook --block-on-critical             # as a git hook
align check --format claude                        # agent-readable output

Flag	Purpose
--base <ref>	Diff <ref>...HEAD instead of the working tree. Use this in CI
--ci	CI mode
--all	Check everything, not just the diff
--advisory	Report, never fail
--block-on-critical	Non-zero exit on a critical conflict
--format	claude, gemini, pi, opencode, text
--resolve <id>:<honored|overridden|context_changed>	Record how you resolved a flagged decision

In CI, always pass --base

Without it, align check diffs the working tree - and a clean CI checkout has nothing staged and nothing uncommitted. It will report "no changes to check" and exit 0 on every pull request forever, which looks exactly like a passing gate.

The check returns unknown rather than a pass when it cannot reach the analysis service. Treat unknown as "unchecked", not as "fine". See the Agent Alignment Layer.

MCP server

align mcp --setup     # writes the config for detected editors
align mcp             # run it directly (stdio)

Manual configuration:

{
  "mcpServers": {
    "align": {
      "command": "align",
      "args": ["mcp"]
    }
  }
}

The CLI server exposes 8 tools with underscored names (search, capture, check_alignment, …). The hosted server exposes 19 with dotted names (search). They are different servers reading different graphs - see AI Assistants (MCP).

Environment variables

Variable	Purpose
ALIGN_ENV	local, preview or prod
ALIGN_GATEWAY_URL	Override the API URL
ALIGN_TOKEN	API token, for non-interactive use
ALIGN_TENANT_ID	Workspace tenant ID

What the CLI does not do

It builds a personal graph. Real-time capture from Slack or Teams, ambient suggestions, cross-team conflict detection and the shared org graph are hosted features - see Plans. Nothing you capture with the CLI is lost when you move to a shared workspace.

Next steps

• AI Assistants (MCP) - wire it into Claude, Cursor or Copilot
• Integrations - hosted connectors vs CLI sources
• Plans - what the shared graph adds