OpenAI Responses API and Agents SDK
by the PREVIZ team
Works today with an API key in a header.
Before you start
https://mcp.aimpreviz.comStreamable HTTP, JSON-RPC 2.0. Your key goes in the Authorization header as a bearer token.You need a PREVIZ account and an API key. Mint the key in Account → API keys and give it a monthly credit cap: an agent holding the key can spend inside that cap and no further, and a paid tool answers with a quote and runs nothing until it is called again with that quote's token. The key is shown once; keep it in an environment variable named PVZ_KEY, which the snippets below read.
A key holds only the access classes its owner gave it, set in Account → API keys → the key's Access: a new key holds read and write, and spend needs a monthly credit cap. tools/list shows only the tools a key holds, and a call to any other is refused with INSUFFICIENT_SCOPE. A key made before access classes existed keeps what it could do then (read, write, spend and live) until its owner sets its access; produce is given, never inherited.
Set it up
- 01
Responses API
Declare PREVIZ as a remote server tool. Leave approval on while you build: a paid PREVIZ tool called without a quoteToken or an idempotencyKey returns a quote and runs nothing, and approval lets you see each call, and whether it carries one, before it is sent. Replace YOUR_MODEL with the id of a model your own account can use; the request is refused until you do.
Pythonimport os from openai import OpenAI client = OpenAI() resp = client.responses.create( model="YOUR_MODEL", tools=[{ "type": "mcp", "server_label": "previz", "server_url": "https://mcp.aimpreviz.com", "headers": {"Authorization": f"Bearer {os.environ['PVZ_KEY']}"}, "require_approval": "always", }], input="List my PREVIZ projects.", ) - 02
Agents SDK
Attach PREVIZ to an agent as a streamable HTTP server.
Pythonimport os from agents import Agent, Runner from agents.mcp import MCPServerStreamableHttp async def main(): async with MCPServerStreamableHttp( name="previz", params={"url": "https://mcp.aimpreviz.com", "headers": {"Authorization": f"Bearer {os.environ['PVZ_KEY']}"}}, ) as previz: agent = Agent(name="Director", mcp_servers=[previz]) result = await Runner.run(agent, "List my PREVIZ projects.") print(result.final_output)
Config formats belong to each client and change with its releases; if a field name above is refused, check the client's own documentation for its current remote-server entry. The address and the header never change.
Test the connection
Two free calls prove the whole path: the key, the address, the client and your projects. Neither spends a credit.
- Ask for your PREVIZ projects. The client calls list_projects and answers with their names and ids. An empty list means the key reaches no project yet, not that the connection failed.
- Ask what a cinematic frame of something would cost in one of them. The client calls plan_quote or generate_frame without a quote token, and answers with a price. Nothing renders and nothing is charged.
- If the client reports 401 or unauthorized, the key is missing, mistyped or revoked; if it reports no tools, the address or the transport is wrong. Fix that before anything else.
What to know
- The Responses API calls PREVIZ from OpenAI's servers with the key you pass; give it a key with a monthly cap.
- Spend with the quote's quoteToken or your own idempotencyKey, never a bare confirm: then a call the framework retries after a timeout joins the first run and is charged once (see Quotes and money).
Then
Once the test passes, the reference and the concepts are where the work is.