haystack

Metorial + Haystack

Uses deepset Haystack with OpenAI GPT-4o to run a pipeline with MCP tool calls via Metorial. The example uses Metorial Search (built-in web search) by default — no dashboard setup needed.

Environment variables

• METORIAL_API_KEY — get one at platform.metorial.com
• OPENAI_API_KEY — from platform.openai.com

Run

cp .env.example .env
pip install -r requirements.txt
python example.py

How it works

This README snippet uses bare await for readability. For a runnable script, see example.py.

import os

from haystack import Pipeline
from haystack.components.generators.chat import OpenAIChatGenerator
from haystack.components.tools import ToolInvoker
from haystack.dataclasses import ChatMessage

# Initialize the Metorial client
from metorial import Metorial, metorial_haystack

metorial = Metorial(api_key=os.getenv("METORIAL_API_KEY"))

# Create a deployment for Metorial Search
deployment = metorial.provider_deployments.create(
    name="Metorial Search",
    provider_id="metorial-search",
)

# Connect and resolve Haystack tools directly
session = await metorial.connect(
    adapter=metorial_haystack(),
    providers=[
        {"provider_deployment_id": deployment.id},
    ],
)

# Tools are used by both the OpenAIChatGenerator (which tells the LLM
# about available tools) and the ToolInvoker (which executes tool calls)
tools = session.tools()
generator = OpenAIChatGenerator(model="gpt-4o", tools=tools)
tool_invoker = ToolInvoker(tools=tools)

# Build the pipeline — connect the generator's output to the tool
# invoker. When GPT-4o requests a tool call, Haystack routes it
# through the invoker which calls Metorial, and the results are
# returned.
pipeline = Pipeline()
pipeline.add_component("generator", generator)
pipeline.add_component("tool_invoker", tool_invoker)
pipeline.connect("generator.replies", "tool_invoker.messages")

# Run the pipeline
messages = [ChatMessage.from_user(
    "Search the web for the latest news about AI agents and summarize the top 3 stories."
)]
result = pipeline.run({"generator": {"messages": messages}})
print(result["tool_invoker"]["tool_messages"])

Adding OAuth providers

To add a provider that requires OAuth (like Slack or GitHub), uncomment the second entry in the providers list and provide your deployment and auth config IDs. See the main README for details on setting up OAuth.