Skip to main content
Every agent uses the same API and the same Summer sign-in. Coding agents use the summer CLI. Chat apps use the Summer MCP. You sign in once, in the browser, and choose Allow with a spend limit.

​
The one step for coding agents

Paste this into Claude Code, Codex, Cursor, Hermes or any agent that can run commands:
Prompt for your agent
The browser opens the Summer sign-in page. You sign in, pick a spend limit and choose Allow. The agent makes its first asset.

​
The summer CLI

One browser sign-in. It renews itself, so the agent never asks again on this computer. Then every Summer API call is one command: Run them as npx -y summer-engine@latest api ..., or as summer api ... once the CLI is installed. The output is JSON when piped or with --json, and a short summary in a terminal. In Windows PowerShell or cmd, put the input in a file and pass --input "@input.json". The CLI’s MCP server has the same calls as tools: summer_api_models, summer_api_estimate, summer_api_run, summer_api_batch and summer_api_job. npx -y summer-engine@latest setup <agent> --yes also installs the summer-api skill, which teaches the agent this whole flow, and the summer-router skill for AI characters in games.

​
ChatGPT and Claude

Chat apps use the Summer MCP. Add this address as a custom connector and sign in to your Summer account once:
Then ask in plain words, for example “Make a low-poly pine tree for my game with the Summer API”. The agent finds the Summer API operations (models.list, models.estimate, models.run) with summer_search and runs them with summer_call. Results go to your Summer library. Step by step for each app: Summer in ChatGPT and Claude. Coding agents can add the same server: Claude Code claude mcp add --transport http summer-cloud https://mcp.summerengine.com/mcp, Codex codex mcp add summer-cloud --url https://mcp.summerengine.com/mcp. Every app you connect shows in Settings > Connected agents & apps with its own spend limit and Revoke. See sign-in and limits.

​
What agents should always do

  • Read models and prices from summer api models or GET /v1/models, never from memory.
  • Price a big batch first with summer api estimate or POST /v1/estimate, and tell the person the total. Then send it as one batch (summer api run --batch, POST /v1/batches), not job by job.
  • Treat queued as accepted: the job starts in order and its price is held. Never send it again.
  • Send a new Idempotency-Key for each new job and the same one when retrying it. The CLI does this for you.
  • Save finished files in the project, and keep the asset id to import the asset again.
  • Keep tokens out of game code, builds and commits.
  • On 402, show the person the error message and stop. no_credits links where to add credits; over_limit links where to raise this app’s limit.