Install Page Systems in an OpenAI Agents SDK agent
Connect the agent to the Page Systems MCP server. Its tools then appear to the model next to your own: list the page types, read a schema, create a page, read the answer.
Tested with openai-agents 0.23.1 (Python) and @openai/agents 0.20.0 (TypeScript).
1. Set your API key
Create a key in the dashboard (it starts with ap_live_ and is shown once) and set it where your agent runs. Never commit it.
export PAGE_SYSTEMS_API_KEY=ap_live_...2. Python
pip install openai-agentsimport os
import urllib.request
from agents import Agent, Runner
from agents.mcp import MCPServerStreamableHttp
# The Page Systems skill: when to send a page and how to use pages well. Fetch it once at startup.
skill = urllib.request.urlopen("https://pagesystems.ai/v1/skill").read().decode()
async def main() -> None:
async with MCPServerStreamableHttp(
name="page-systems",
params={
"url": "https://pagesystems.ai/mcp",
"headers": {"Authorization": f"Bearer {os.environ['PAGE_SYSTEMS_API_KEY']}"},
},
cache_tools_list=True,
) as page_systems:
agent = Agent(name="Assistant", instructions=your_instructions + "\n\n" + skill, mcp_servers=[page_systems])
result = await Runner.run(agent, user_message)
print(result.final_output)3. TypeScript
npm install @openai/agentsimport { Agent, MCPServerStreamableHttp, run } from '@openai/agents';
const skill = await fetch('https://pagesystems.ai/v1/skill').then((r) => r.text()); // once, at startup
const pageSystems = new MCPServerStreamableHttp({
name: 'page-systems',
url: 'https://pagesystems.ai/mcp',
requestInit: { headers: { Authorization: `Bearer ${process.env.PAGE_SYSTEMS_API_KEY}` } },
cacheToolsList: true,
});
await pageSystems.connect();
try {
const agent = new Agent({ name: 'Assistant', instructions: [yourInstructions, skill].join('\n\n'), mcpServers: [pageSystems] });
const result = await run(agent, userMessage);
console.log(result.finalOutput);
} finally {
await pageSystems.close();
}4. Keep the conversation
The answer arrives in a later turn, so pass the history back in: result.to_input_list() in Python, result.history in TypeScript, plus the new user message. The model then calls get_page_result with the page id it already has.
5. Check it works
Ask your agent something a page answers better than chat, for example "Ask Sam which of 9:00, 11:00 or 14:00 next Tuesday works for a call." It should send a link. Open it, pick a time, submit, then ask the agent what Sam picked: it reads the answer with get_page_result.