Agents API

List agent metadata (tier, category, trust ranking, accessibility) from the control plane. Day-to-day invocation uses the CLI or MCP, not a typed SDK.

Truth note (bs-8328): The public control-plane route is GET /api/v1/agents (metadata). It does not return fake marketplace stats or ship a published @bootspring/sdk. Agent invoke for product workflows is CLI/MCP; advanced hosted invoke lives on the engine when configured.

Control-plane endpoints#

Preferred invoke path (Live)#

List agents (REST)#

Example#

Response shape#

Notes:

• Trust ranking is evidence-first (ratingsDownloadsWeight: 0 until live outcome receipts exist).
• accessible reflects the caller's plan tier vs agent tier gates.
• Do not treat marketing counts (e.g. "36 agents") as a guarantee that every agent is production-routed for every plan.

Agent context#

Returns the agent context document used by tooling. Requires authentication. Prefer CLI/MCP when you need interactive agent runs.

Hosted engine invoke (advanced)#

The hosted engine exposes agent automation routes (including invoke) for server integrations. Prefer CLI/MCP for product workflows. When calling the engine directly:

• Authenticate with a dashboard API key
• Send a task/prompt payload matching the engine contract
• Expect rate limits and tier gates

Do not copy fictional SDK samples such as import { Bootspring } from '@bootspring/sdk' — that package is roadmap-only and is not published.

Errors#

Related#

• Entitlements API — feature gates and plan resolve
• Track API — usage metering for invocations
• MCP Server — assistant tool surface
• CLI reference — primary operator interface