{"name":"Runtype MCP Server","version":"0.1.0","protocol":"mcp","protocolVersion":"2025-11-25","description":"Model Context Protocol server for Runtype API","transport":"streamable-http","capabilities":{"tools":{},"resources":{},"prompts":{}},"instructions":"When the user wants to build an AI product, design or run a flow, configure a surface, embed a Persona chat widget, or manage agents, records, logs, or runtime tools on Runtype, use this server. For any build or design task, call `get_build_instructions` (tasks: `build-product`, `generate-flow`, `explain-capabilities`) BEFORE any `create_*` mutation — it returns the platform rules, schemas, step types, and variable-system guidance for the task. Use `get_platform_documentation` for schema, type-definition, tool-catalog, embed, SDK, or dashboard-link lookups, and `search_documentation` for prose/how-to questions answered from the docs corpus. Best practices: set an explicit, enabled model on every agent and prompt step (claude-sonnet-4-6 by default, or claude-haiku-4-5 for speed) — never omit it or rely on account defaults; validate with `validate_flow`/`validate_product` before creating, and test with `execute_agent`/`run_flow` before sharing a result; for anything meant for repeated use, create an eval suite (`create_eval_suite`) alongside it and re-run it (`run_eval_suite`) after every change — and when troubleshooting reveals a real failure, capture it as a regression case (`add_eval_cases`) before fixing. Prompts (`build-product`, `generate-flow`, `explain-capabilities`) expose the same guidance for clients that surface prompts directly to users."}