MCP — use from Claude & agents

GetMeCompute speaks the Model Context Protocol: connect Claude Code, Claude Desktop, Cursor, or any MCP client and your AI assistant can price, submit, and collect video jobs by itself — you just describe what you want.

Connect in one command

The hosted MCP endpoint lives at https://getmecompute.com/api/mcp — nothing to install. Authenticate with an API key from Dashboard → API Keys.

Claude Code

claude mcp add getmecompute --transport http https://getmecompute.com/api/mcp \
  --header "Authorization: Bearer gmc_live_XXXXXXXX"

Cursor / other clients (mcpServers JSON)

{
  "mcpServers": {
    "getmecompute": {
      "url": "https://getmecompute.com/api/mcp",
      "headers": { "Authorization": "Bearer gmc_live_XXXXXXXX" }
    }
  }
}

Test the connection first: use a sandbox key (gmc_test_…, created from the same dashboard page). The agent goes through the entire flow — pricing, submission, waiting, collecting the result — but sandbox jobs are simulated test runs: no real video is generated, and each completes with a stock sample video within about a minute. Swap in a gmc_live_ key when you're ready for real videos.

What the agent can do

Tool	What it does
list_models	Active models, capabilities, and the full price matrix.
estimate_cost	Price a job in USD without creating it.
create_video_job	Submit a job — returns `job_id` and the charged cost immediately. Supports an idempotency key so agent retries can never double-bill.
wait_for_job	Poll until the video is done and return the download link.
get_job / list_jobs	Status, results, and history.
whoami	Verify which account the key belongs to.

Tool descriptions carry the guardrails agents need: costs are stated up front, renders are flagged as taking minutes, and read-only tools are annotated so clients can call them freely. All validation, pricing, and rate limiting happen server-side in the same REST API — the agent can never set its own price.

A typical conversation

You, in Claude Code

> Take https://example.com/product.jpg and make a 10s 720p clip of it
> rotating slowly on a marble surface. Tell me the price first.

# Claude calls: list_models → estimate_cost ($0.20) → asks you to confirm
# → create_video_job → wait_for_job → hands you the MP4 link.

Run it locally instead

A self-hosted stdio server (@getmecompute/mcp on npm) is coming soon — it adds an upload_image tool that sends images straight from your machine. Until then, the hosted endpoint covers everything with publicly hosted images (or upload via the REST uploads endpoint and pass the returned URL).

Building something else?

The MCP tools are a thin layer over the public API — the job tool exposes the core advanced parameters (seed, sample_steps, negative_prompt, generate_audio), while the full set is REST-only. For GPT Actions, custom agents, or SDK generation, point your tooling at the machine-readable contract: GET https://getmecompute.com/api/v1/openapi.json.

Webhooks & automation