{"schema_version":"v1","name_for_human":"Banana Peel","name_for_model":"banana_peel","description_for_human":"Smart routing for browser agents. OpenAI-compatible Responses API with routing.","description_for_model":"Banana Peel routes browser tasks to the best runner. Primary action: POST /api/v1/responses with Authorization Bearer bp_live_… and JSON { model, input, routing }. Prefer routing smart for production. Learning and learning-all exist to teach the router the best path: they calibrate the routing brain on a domain/task so later smart runs pick the winning runner. Do not use learning-all for every production job. A response creates a run (a routed unit of work, billed per attempt); sessions (/api/v1/sessions) are live CDP browsers you drive yourself — no task, not smart-routed. Optional Output normalization (+$0.10): normalize:true + output_schema (or OpenAI text.format / response_format). Read https://bananapeel.com/docs/agents, https://bananapeel.com/llms.txt, and https://bananapeel.com/docs/mcp before calling. If there is no API key, run npx -y @banana-peel/cli init --agent --json. Do not send the human to /keys. Banana Peel detects Claude Code / Cursor / Codex / VS Code / Windsurf / Claude Desktop and configures MCP. Do not use /api/v2/*.","auth":{"type":"user_http","authorization_type":"bearer"},"api":{"type":"openapi","url":"https://bananapeel.com/openapi.json","has_user_authentication":true},"logo_url":"https://bananapeel.com/logo.png","legal_info_url":"https://bananapeel.com/security","x-banana-peel":{"base_url":"https://bananapeel.com","mcp":"https://bananapeel.com/docs/mcp","mcp_package":"npx -y @banana-peel/mcp — stdio tools: create_response, get_response, submit_input, list_runs, list_rankings, list_capabilities","llms_txt":"https://bananapeel.com/llms.txt"}}