agentproto models

agentproto models [adapter] [--json]

Lists every model declared by installed agent adapters, annotating each with its provider and whether a provider API key is configured (✓ runnable, or ✗ missing key). Prices — when known — come from @agentproto/model-catalog ($ per 1M input/output tokens).

If no installed adapter declares a model list, the command prints no adapters installed — try: agentproto install claude-code.

The point: --model anthropic/claude-opus-4-8 fails at runtime with "no ANTHROPIC_API_KEY" if you haven't set a key. This surfaces the issue up front so you pick a model your environment is ready for.

FlagsCopy Anchor Link

OutputCopy Anchor Link

Without --json, prints one adapter block per installed adapter with its status, then one model per line:

mastra-agent  (installed)
  ✓ openrouter/z-ai/glm-5.2                 openrouter $0/$0 per 1M
  ✓ openrouter/deepseek/deepseek-v4-pro     openrouter $0/$0 per 1M
  ✗ openai/gpt-5                            openai     $?/$? per 1M

✗ = no provider key. Set one:
    agentproto auth provider set openai <api-key>

With --json, emits (plus contextWindow / maxOutput per model, when the live-synced CONTEXT_WINDOWS table knows them):

{
  "adapters": [
    {
      "adapter": "mastra-agent",
      "status": "installed",
      "models": [
        {
          "id": "openrouter/z-ai/glm-5.2",
          "provider": "openrouter",
          "runnable": true,
          "inputPer1M": null,
          "outputPer1M": null,
          "contextWindow": 1000000,
          "maxOutput": 128000
        },
        {
          "id": "openrouter/deepseek/deepseek-v4-pro",
          "provider": "openrouter",
          "runnable": true,
          "inputPer1M": null,
          "outputPer1M": null
        },
        {
          "id": "openai/gpt-5",
          "provider": "openai",
          "runnable": false,
          "inputPer1M": null,
          "outputPer1M": null
        }
      ]
    }
  ]
}

Provider key sourcesCopy Anchor Link

A model is marked runnable (✓) when its provider has a key available from either:

1. The current process environment — ANTHROPIC_API_KEY, OPENAI_API_KEY, OPENROUTER_API_KEY, etc.
2. The stored provider key file (agentproto auth provider set …).

catalog_models (MCP) — the route/vendor/profile-aware catalogCopy Anchor Link

agentproto models lists what each adapter declares. For the broader view — every model this host can reach, widened beyond any one adapter's list via OpenRouter / Requesty / HuggingFace routing — the daemon exposes the read-only catalog_models MCP tool (backed by GET /catalog/models):

• entries are keyed by vendor / product and carry one entry per route (endpoint rail: anthropic, openrouter, requesty, …);
• each route has a profile-aware runnable flag — true only when an auth profile eligible for that (adapter × route) is actually configured on the host, so it answers "what is spawnable right now", not just "what key env var exists";
• filter with adapter, vendor, route, and runnableOnly.

Use it before agent_start to pick a (model, route) pair the host can run, and alongside the access config axis (named auth profiles) that decides which profile pays.

ExamplesCopy Anchor Link

# All adapters, all models
agentproto models

# One adapter in detail
agentproto models mastra-agent

# Machine-readable
agentproto models --json | jq '.adapters[].models[] | select(.runnable) | .id'

See alsoCopy Anchor Link

• auth.md — set provider keys with agentproto auth provider set
• run.md — pass --model to a one-shot turn
• chat.md — pass --model to an interactive session
• concepts/adapters.md — how adapters declare models