CLI Reference
Applies to SDK 0.16+ · Last updated: 2026-06-12
The agentnode CLI ships with the SDK (pip install agentnode-sdk). Commands are grouped by task below. Running agentnode with no command shows a status dashboard; global flags --version and --no-color apply to every command.
The five you will use most:
terminal
$ agentnode setup # one-time interactive setup
$ agentnode search "pdf" # find a capability
$ agentnode install pdf-reader-pack
$ agentnode run pdf-reader-pack --input '{"file_path":"report.pdf"}'
$ agentnode auth status # check configured providersNew here? Start with the Quick Start — this page is the full reference.
Setup & configuration
| Command | Description |
|---|---|
| agentnode setup | Interactive wizard covering all first-class settings as multiple-choice prompts with recommended defaults: install behavior, trust level, permissions, guard posture, credentials, sandbox and host_trust_policy, plus an advanced gate. Skippable — accepting the recommendations reproduces today's config. |
| agentnode config list | Show all settings with their descriptions and allowed values. |
| agentnode config get <key> | Read one config value. |
| agentnode config set <key> <value> | Set a config value, e.g. config set llm.default_provider ollama. |
| agentnode config set sandbox.host_trust_policy <default|curated_only|none> | Choose which trust tiers may run directly on your host; stricter values sandbox trusted/curated (fail-closed, reinstall may be needed). |
| agentnode doctor [--fix] [--json] | Diagnose your setup and detect missing capabilities. --fix auto-installs suggestions. |
| agentnode reset | Reset the local configuration to defaults. |
Find, install & run
| Command | Description |
|---|---|
| agentnode search <query> | Full-text search of the registry. |
| agentnode discover [--trending] [--new] [--category <c>] [--type toolpack|agent] | Browse trending, newly published, or categorized packages. |
| agentnode resolve <capability...> [--framework <f>] [--json] | Rank packages for one or more capability IDs using the scoring engine. |
| agentnode recommend [--json] | Suggest packages that complement what you already have installed. |
| agentnode install <slug> [--version <v>] [--yes] | Install a package: policy check, download, SHA-256 verify, lockfile update. |
| agentnode run <slug | task> [--input <json>] [--file <path>] [--explain] [--dry-run] [--raw] [--json] | Run an installed tool (by slug) or a natural-language task. Runs in subprocess isolation. |
| agentnode remove <slug> [--yes] | Remove an installed package from the lockfile. |
Inspect & history
| Command | Description |
|---|---|
| agentnode inspect <slug> [--json] | Security report for an installed package: trust, permissions, risk, signature, policy/guard preview. |
| agentnode capabilities | List installed packages. capabilities show <name> prints details for one. |
| agentnode skill list | List installed skills. skill show <slug> [--render KEY=VALUE] [--raw] renders a skill prompt. |
| agentnode logs [run_id] [--limit <n>] [--json] | Show agent run logs, or details for one run_id. |
| agentnode audit [--limit <n>] [--event <e>] [--slug <s>] [--action allow|deny|prompt] [--tool <t>] [--json] | Read the Guard/policy decision trail from ~/.agentnode/audit.jsonl. |
Trust & safety
| Command | Description |
|---|---|
| agentnode auth set <provider> [--from-env <VAR>] | Store a provider credential (entered hidden). --from-env imports an existing env var. |
| agentnode auth list | Show configured credentials (no secrets shown). |
| agentnode auth status | Show every provider and its effective source (env vs stored). |
| agentnode auth test <provider> | Validate a key via a free endpoint; for Ollama, a localhost reachability check. |
| agentnode auth remove <provider> | Delete a stored credential. |
| agentnode guard policy|status [--json] | Show the resolved guard policy, or the recent guard activity summary. |
| agentnode guard set <action> <allow|prompt|deny> [--tool <key>] | Set a guard action policy, globally or per tool. |
| agentnode guard unset [<action>] --tool <key> | Remove a per-tool guard override (one action, or all for that tool). |
| agentnode guard check <slug/tool> [--action <a>] [--json] | Dry-run the guard decision for a tool without executing it. |
| agentnode guard reset | Reset guard policies to defaults. |
| agentnode lock seal [--force] | Compute integrity hashes for all lockfile entries. |
| agentnode lock verify [--strict] [--online] [--json] | Verify lockfile integrity; --online checks publisher key status against the registry. |
| agentnode sandbox pull | Explicitly pull the digest-pinned sandbox image (no auto-pull). |
| agentnode sandbox doctor [<slug>] [--json] | Diagnose sandbox readiness (now host-trust-policy aware), optionally for one package. |
| agentnode sandbox status | One-line sandbox readiness check. |
Build & publish
| Command | Description |
|---|---|
| agentnode init [name] [--type local|api|file|agent|skill|mcp] | Scaffold a new package from a template. --type mcp scaffolds an MCP server listing. |
| agentnode validate [path] | Validate a manifest offline; check cases and predict the maximum achievable tier. |
| agentnode record-cases [path] [--strict] | Record VCR cassettes for API verification cases. --strict fails on detected secrets. |
| agentnode verify-local [path] | Run the full verification pipeline locally (install, import, smoke, tests, scoring). |
| agentnode publish [path] [--dry-run] [--skip-validate] [--token <key>] [--yes] | Publish a package to the registry. --yes is required for non-interactive/CI publish. |
| agentnode mcp doctor <slug> [--skip-start] [--json] | Check an MCP server's readiness. |
| agentnode mcp verify [path] [--test] [--json] | Verify an MCP agentnode.yaml manifest; --test runs a protocol test. |
| agentnode mcp submit [path] [--test] [--token <key>] [--dry-run] | Submit an MCP server for catalog review. |
| agentnode mcp status <submission_id> [--token <key>] [--json] | Check the status of an MCP submission. |
| agentnode mcp ownership challenge <package> --registry npm|pypi [--token <key>] [--json] | Issue a one-time publish-challenge to prove package ownership. |
| agentnode mcp ownership verify <package> --registry npm|pypi [--token <key>] [--json] | Verify a pending publish-challenge (checks the latest published version). |
Full publishing walkthrough: Publishing Guide and Package Verification. For the complete MCP verification, ownership, submission, and review workflow, see Publish an MCP server.
Advanced
| Command | Description |
|---|---|
| agentnode serve | Run AgentNode itself as an MCP server over stdio. |
Related
- Quick Start — the guided first-run path.
- Credentials & Connectors — details behind
agentnode auth. - Execution Sandbox — details behind
agentnode sandbox. - AgentNode Guard — details behind
agentnode guardandagentnode audit. - Python SDK — the programmatic equivalent of these commands.