Hedra's developer platform: API, SDK, CLI and MCP vs Sume

Hedra's blog post of August 4, 2026 says its models are now a developer platform with four surfaces: a raw API, native Python and TypeScript SDKs, a CLI and MCP. Sume covers the same ground for avatar video with a REST API, a TypeScript SDK (@sume-com/sdk) and a hosted MCP server at https://mcp.sume.com/mcp, with idempotency keys and signed webhooks on both sides.

What does Hedra say each surface is for?

The post gives a rule of thumb: call the API directly for quick, agent-driven queries; use the SDKs when you are shipping code; use the CLI or MCP to hand inference to an agent. It describes the CLI as token-efficient for sandboxes and MCP as self-describing for desktop and chat agents.

It also lists preflight cost estimates, live SSE progress for every generation, durable idempotent jobs, signed webhooks configurable in the dashboard or per request, and a move from expiring credits to dollars.

What does an avatar call look like on Sume?

Create the avatar once with POST /v1/avatar-1.0/generate, then render with POST /v1/avatar-1.0/talking-video, which takes avatar_handle and one of script or video_inputs. The script is accepted when Sume estimates 4-60 seconds. Poll GET /v1/jobs/{id}/status, or pass a webhook_url and handle terminal callbacks.

curl -X POST https://api.sume.com/v1/avatar-1.0/talking-video \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: spokesperson-001" \
  -d '{
    "avatar_handle": "product_host",
    "script": "Meet the new dashboard. Setup takes two minutes.",
    "quality": "plus",
    "mode": "async"
  }'

Which surface should an agent use on Sume?

For an agent in a chat client, connect the hosted MCP server over OAuth or an API key. The MCP docs list read tools such as avatars_list and avatar-videos_get and paid tools such as avatar-videos_create, and say to use dry_run or generation_admission_preview before expensive bursts.

For code in a repository, use the SDK or plain HTTP. The SDK docs say it adds waitForJob, uploadFile and verifyWebhook on top of the generated operations.