OpenAI

jitprompt is a remote MCP server, so OpenAI models can use it two ways:

• Responses API — OpenAI’s servers call jitprompt’s tools for you; your code makes one request.
• Agents SDK — your process connects to jitprompt and runs the agent loop.

Both need an API key for the namespace (namespace → API keys). Keep it in an environment variable, not in code.

Responses API

Add jitprompt as an mcp tool. Pass the namespace key in authorization; OpenAI sends it as the bearer token on each call to jitprompt and does not store it, so send it with every request.

import os
from openai import OpenAI

client = OpenAI()

resp = client.responses.create(
    model="gpt-6-astra",
    instructions=(
        "You are a support agent. Your operating instructions are in the jitprompt tools: "
        "call get_toc first, and search_instructions before each new task. Follow what they return."
    ),
    tools=[{
        "type": "mcp",
        "server_label": "jitprompt",
        "server_description": "The company's operating instructions for this agent.",
        "server_url": "https://app.jitprompt.com/mcp",
        "authorization": os.environ["JITPROMPT_KEY"],
        "allowed_tools": ["get_toc", "search_instructions", "get_section"],
        "require_approval": "never",
    }],
    input="Can I get a refund on my annual plan?",
)
print(resp.output_text)

• require_approval: "never" lets the model call the tools without a round trip to you; all three are read-only.
• allowed_tools limits the import to jitprompt’s three tools.
• The same request with curl:

curl https://api.openai.com/v1/responses \
  -H "Content-Type: application/json" -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "gpt-6-astra",
    "tools": [{
      "type": "mcp",
      "server_label": "jitprompt",
      "server_url": "https://app.jitprompt.com/mcp",
      "authorization": "'"$JITPROMPT_KEY"'",
      "require_approval": "never"
    }],
    "input": "Can I get a refund on my annual plan?"
  }'

The Authorization header there is your OpenAI key; the jitprompt key goes in the tool’s authorization field.

Agents SDK

pip install openai-agents

import asyncio
import os

from agents import Agent, Runner
from agents.mcp import MCPServerStreamableHttp


async def main():
    async with MCPServerStreamableHttp(
        name="jitprompt",
        params={
            "url": "https://app.jitprompt.com/mcp",
            "headers": {"Authorization": f"Bearer {os.environ['JITPROMPT_KEY']}"},
            "timeout": 10,
        },
        cache_tools_list=True,   # jitprompt's tool list doesn't change between runs
        max_retry_attempts=3,
    ) as jitprompt:
        agent = Agent(
            name="Support agent",
            instructions=(
                "Your operating instructions are in the jitprompt tools: call get_toc first, and "
                "search_instructions before each new task. Follow what they return."
            ),
            mcp_servers=[jitprompt],
        )
        result = await Runner.run(agent, "Can I get a refund on my annual plan?")
        print(result.final_output)


asyncio.run(main())

tool_filter on MCPServerStreamableHttp can narrow the tools further, for example to search_instructions only.

See MCP tools for what each tool returns, and Writing prompts for agents for the system-prompt line that makes agents search at the right moments.