--- title: CLI command reference description: Every command, flag and exit code in the published looot CLI (npm looot 1.1.0), plus env vars and the config file. --- `looot` is published on npm as `looot`, currently version 1.1.0. Install it with `npm install -g looot@1.1.0` (see [Install](/get-started/install)); `looot doctor` checks the Node version and compares your installed version against the one published on npm. ```txt looot [options] ``` `looot help` prints a short summary. `looot help --all` prints every command with all its flags. `looot help ` details one command. ## Output On a terminal, output is a short human summary by default. Piped or redirected, it's JSON. Add `--format json`, `--format jsonl` or `--format human` after the command to choose explicitly either way; `--format` only works right after the command. `jsonl` splits a top-level array result into one JSON object per line; it is not network streaming, and every other result stays one line. Human output redacts secret-looking fields. ## Env and config `--url` beats `LOOOT_API_URL` beats the saved `~/.config/looot/config.json` beats `https://api.looot.ai`. `LOOOT_TOKEN` beats the saved login. There is no `--token` flag: a secret never sits in argv or shell history, so use `looot login` or set `LOOOT_TOKEN`. `looot login` saves the token to `~/.config/looot/config.json`, readable only by you. ## Exit codes | Code | Meaning | | --- | --- | | `0` | Success. | | `1` | Any failure, including a run that settles `failed` or `blocked`. | | `130` | Ctrl+C. | ## Commands ### login ```txt looot login [--token-stdin] ``` Opens `https://looot.ai/cli/approve` in your browser with a code, waits up to 10 minutes for you to approve it there, and saves the credential. The raw token never appears in the URL. `--token-stdin` instead reads a token from stdin (opening `/connect` and verifying it once, no polling): `echo "$TOKEN" | looot login --token-stdin`. ### logout ```txt looot logout ``` Revokes the saved token on the server, then deletes the local config file either way. A failed revoke leaves the token valid until it expires or is revoked in Settings. Only the token `login` saved is revoked, never a `LOOOT_TOKEN` from the environment (which may be shared with CI). ### whoami ```txt looot whoami ``` Prints the organization, role and granted scopes for the signed-in token, plus its last 4 characters only. ### token ```txt looot token --reveal ``` Prints the saved (or resolved) token to stdout, for a script: `export LOOOT_TOKEN="$(looot token --reveal)"`. Without `--reveal` it refuses, on a terminal and in a pipe alike, since an agent's shell is always a pipe. With `--reveal` it warns on stderr that the token itself is on stdout. The older `--yes-print` flag still works. ### init ```txt looot init [--yes] [--dry-run] [--embed] [--project] ``` Logs in if needed, detects Claude Code, Cursor and Codex, and offers to write the looot MCP entry into each one it finds: a credential reference by default (`--embed` writes the token itself into the file instead), then prints two commands to try and your balance. `--yes` skips the per-client confirmation prompt. `--dry-run` asks nothing, logs in to nothing, calls nothing, writes nothing, and only prints the `looot` entry it would have written. `--project` writes to the project-local agent config; without it, `init` writes the user-level one. ### doctor ```txt looot doctor [--project] ``` Checks the Node version, gateway reachability, token validity, clock skew, which agent clients it detects, and this install's version against the one published on npm. ### search ```txt looot search "text" [--limit N] [--provider ID] [--category ID] ``` A human table of matching endpoints: id, provider, price per call. An alias for `catalog search --query "text"`. Works signed out, reading the public catalog; see [Browse without an account](/get-started/browse). ### inspect ```txt looot inspect ``` The exact input and output schema, price formula, estimated maximum cost (when signed in) and provider identity for one endpoint, before you run it. Works signed out too, on the public catalog. ### catalog ```txt looot catalog endpoints [--provider ID] [--category ID] [--platform ID] [--capability ID] [--limit 1..5000] [--cursor CURSOR] [--keyless true|false] [--mock true|false] [--facets true|false] [--max-price-micros N] looot catalog search --query TEXT [same filters as endpoints] looot catalog list | inspect | compare ... ``` Published operation metadata. Provider-only prices are not quotes, and execution availability is a separate check from listing. `catalog list`, `catalog inspect` and `catalog compare` read the public catalog and work signed out. ### run ```txt looot run --input '{"...": "..."}' [--idempotency-key KEY] [--wait[=seconds]] ``` Starts, or replays, a run. An omitted `--idempotency-key` is generated on the CLI itself, printed to stderr, and safe to reuse on a retry (see [Idempotency](/concepts/idempotency)). `--wait` blocks and prints the finished run inline, not the queued or running one. It takes whole seconds, 0 to 60: `0` returns at once, the default is 20 when `--wait` is passed bare, and a value above 60 waits 60 and says so on stderr. :::note `looot run job:` works in 1.1.0: the CLI sends the id as written and the gateway picks the provider (see [Jobs](/concepts/jobs)). `--fallback` is not in 1.1.0 yet; it comes with the next CLI release. Use the MCP [`run`](/mcp-tools/run) tool or `POST /v1/runs` today for fallback. See [Fallback](/concepts/fallback). ::: ### runs ```txt looot runs get looot runs list [--status ...] [--capability ...] [--limit N] [--cursor CURSOR] looot runs cancel looot runs attempts looot runs evidence ``` `runs evidence` prints the same attempts JSON as `runs attempts`; in human output it also adds the input fingerprint, provider request and receipt ids, and quote and ledger entries. ### balance ```txt looot balance ``` Available and reserved balance for this workspace, and `topUpLink.minimumUsd` for the minimum top-up. ### ledger ```txt looot ledger list [--limit 1..50] [--cursor CURSOR] ``` Lists ledger entries for this workspace. Needs the `usage.read` scope. ### mcp ```txt looot mcp install --client claude-code|cursor|codex|generic [--auth-mode env|embedded] [--token-env NAME] [--show-token] ``` Writes an offline Streamable HTTP client config pointed at `$LOOOT_API_URL/mcp`. Prefer `looot init`, which does this for you across every detected client. HTTPS is required except on loopback. Embedded mode prints a placeholder where your token goes, unless you pass `--show-token` (which warns on stderr). ## Not in 1.1.0 yet The live gateway already accepts these; the next CLI release adds the flags. Use MCP or REST until then. | Feature | Use instead | | --- | --- | | `--fallback` on `run` | MCP `run` tool, or `fallback` in the `POST /v1/runs` body. | | `--prefer` on `search`/`discover` | MCP [`search_catalog`](/mcp-tools/search-catalog)/`discover`, or `?prefer=` on the REST routes. | | `looot catalog overview` | MCP [`catalog_overview`](/mcp-tools/catalog-overview), or `GET /v1/catalog/overview`. | See [Fallback](/concepts/fallback) and [Jobs](/concepts/jobs) for what these do.