openai

Predactiv MCP — OpenAI GPT AI Agent

This sample shows how to connect an AI agent to the Predactiv MCP server using OpenAI and its Responses API.

Unlike a client-side agent loop, this sample uses OpenAI's native MCP connector: the MCP server definition is handed to the Responses API, and OpenAI's infrastructure connects to the server, discovers its tools, and runs the tool-calling loop server-side. Your code makes a single responses.create call and reads back the final answer.

How it works

1. Authenticate — oauth_token_lib.get_bearer_token() exchanges your CLIENT_ID and CLIENT_SECRET for an OAuth2 bearer token (client-credentials flow).
2. Connect & discover tools — the request declares the Predactiv server as an mcp tool (server_label, server_url, authorization). OpenAI connects to the server and discovers its tools dynamically. No tools are hardcoded.
3. Run the agent — GPT connects to the MCP server, calls its tools as needed, and returns a final response. require_approval: "never" lets the model call the tools without a manual approval round-trip.

The Predactiv MCP tools expose the data platform: audiences, datasets, datasources, destinations, filters/predictors, and insights (predactiv_* tool families).

Prerequisites

• Python 3.9+
• A Predactiv account with API credentials (CLIENT_ID and CLIENT_SECRET). Contact Predactiv if you don't have these.
• An OpenAI API key — this sample uses gpt-5.4-mini as the agent's LLM. Set it as the OPENAI_API_KEY environment variable.

Two separate credentials. OPENAI_API_KEY authenticates your calls to OpenAI. CLIENT_ID / CLIENT_SECRET authenticate to the Predactiv MCP server via oauth_token_lib. Both must be set.

Setup

From this openai/ directory:

# (recommended) create and activate a virtual environment
python -m venv .venv
source .venv/bin/activate        # on Windows: .venv\Scripts\activate

# install dependencies (also installs the shared oauth-token-lib package)
pip install -r requirements.txt

Configuration

Set the following environment variables:

export CLIENT_ID=your-predactiv-client-id
export CLIENT_SECRET=your-predactiv-client-secret
export OPENAI_API_KEY=your-openai-api-key

Run

python agent.py

The sample's default goal is:

Get list of my audiences from the Predactiv MCP server and return it to me.

Edit the GOAL constant in agent.py to ask the agent to do something else with your Predactiv data. The model is set in the responses.create call (model="gpt-5.4-mini") — swap it for any model your key can access.

Troubleshooting

• Responses.create() got an unexpected keyword argument 'messages' — The Responses API uses input, not messages (that's the Chat Completions API). Pass input=GOAL.
• 401 Unauthorized from the MCP server — The authorization field expects the raw access token; OpenAI adds the Authorization: Bearer … header itself. Don't prefix the token with Bearer (a double prefix causes a 401). If it persists, verify your credentials with python ../oauth-token-lib/oauth_token_lib.py.
• output_text is empty — The model likely stopped on an mcp_approval_request without calling the tool. Set require_approval: "never" on the MCP tool. The sample also dumps response.output items when there's no text, so you can see what came back.
• OpenAI authentication errors — Make sure OPENAI_API_KEY is set and valid.

Learn more

• Predactiv MCP server: https://mcp.predactiv.com
• OpenAI Responses API: https://platform.openai.com/docs/api-reference/responses
• Model Context Protocol: https://modelcontextprotocol.io