Tabular Data Processing with Prior Labs MCP

!pip install -q haystack-ai anthropic-haystack mcp-haystack trafilatura firecrawl-haystack

import os
from getpass import getpass

if "ANTHROPIC_API_KEY" not in os.environ:
    os.environ["ANTHROPIC_API_KEY"] = getpass("Enter your Anthropic API key: ")
if "FIRECRAWL_API_KEY" not in os.environ:
    os.environ["FIRECRAWL_API_KEY"] = getpass("Enter your Firecrawl API key: ")
if "PRIOR_LABS_MCP_TOKEN" not in os.environ:
    os.environ["PRIOR_LABS_MCP_TOKEN"] = getpass("Enter your Prior Labs MCP token: ")

from haystack_integrations.components.generators.anthropic import AnthropicChatGenerator
from haystack_integrations.components.fetchers.firecrawl import FirecrawlCrawler
from haystack.components.builders import ChatPromptBuilder
from haystack.dataclasses import ChatMessage
from haystack import Pipeline


pipeline = Pipeline()
pipeline.add_component("fetcher", FirecrawlCrawler())
pipeline.add_component(
    "prompt_builder", 
    ChatPromptBuilder(
        template=[
            ChatMessage.from_system(
                "You are a hotel booking assistant. Extract the relevant information from the text and summarize it in a concise manner, as a Markdown table with the following columns: \n- Hotel name\n- Location\n- Distance from the target\n- Price per night\n- Room type\n- User rating\n- Location rating\n- Number of reviews\n- Number and types of beds\n- Amenities\n- Room size\n- Whether breakfast is included\n- Public transport availability\n\nIf any of the information is missing, leave the corresponding cell empty. Always return all the hotels from the input, even if some of the information is missing."
            ),
            ChatMessage.from_user(
                "{% for doc in documents %}{{ doc.content }}\n{% endfor %}"
            ),
        ],
        required_variables=["documents"],
    ),
)
pipeline.add_component(
    "summarizer", 
    AnthropicChatGenerator(
        model="claude-opus-4-6",
        generation_kwargs={
            "max_tokens": 10000,
        }
    ),
)

pipeline.connect("fetcher.documents", "prompt_builder.documents")
pipeline.connect("prompt_builder.prompt", "summarizer.messages")

output = pipeline.run({
    "fetcher": {
        "urls": [
            "https://www.booking.com/searchresults.html?ss=Berlin%2C+Berlin+Federal+State%2C+Germany&efdco=1&label=gen173nr-10CAEoggI46AdIM1gEaLYBiAEBmAEzuAEHyAEM2AED6AEB-AEBiAIBqAIBuAKQgrTOBsACAdICJDllMDgyNTA1LTdlYmMtNDJkNi04NzFmLWUyYTRkNWM1NjM3ZtgCAeACAQ&aid=304142&lang=en-us&sb=1&src_elem=sb&src=index&dest_id=-1746443&dest_type=city&ac_position=0&ac_click_type=b&ac_langcode=xu&ac_suggestion_list_length=5&search_selected=true&search_pageview_id=0da2508885520c7d&ac_meta=GhAwZGEyNTA4ODg1NTIwYzdkIAAoATICeHU6BmJlcmxpbg%3D%3D&checkin=2026-05-05&checkout=2026-05-08&group_adults=1&no_rooms=1&group_children=0",
        ]
    }
})

from IPython.display import Markdown, display

display(Markdown(output["summarizer"]["replies"][-1].text))

from haystack_integrations.tools.mcp import MCPToolset, StreamableHttpServerInfo
from haystack.utils import Secret

# Get the server info from environment variables and create the toolset
# based on the tools exposed by the MCP server.
server_info = StreamableHttpServerInfo(
    url="https://api.priorlabs.ai/mcp/server",
    token=Secret.from_env_var("PRIOR_LABS_MCP_TOKEN"),
)
toolset = MCPToolset(server_info=server_info, eager_connect=True)

for tool in toolset.tools:
    print(f"{tool.name}: {tool.description}")

from haystack.components.agents import Agent

llm = AnthropicChatGenerator(
    model="claude-opus-4-6",
    generation_kwargs={
        "max_tokens": 10000,
    }
)

pipeline.add_component(
    "agent",
    Agent(
        chat_generator=llm,
        tools=toolset,
        user_prompt="""{% message role="user" %}
You are a data scientist working with tabular data with a specialization in travel and hospitality. You have access to a set of tools that you should use to predict the missing attributes of hotels based on the information provided in the input table.
Before you use any of the tools, make sure to prepare the hotel data appropriately, like you would do for any tabular dataset, and interpret the results to create a human-friendly table with a structure similar to the input table.
Return only a Markdown-formatted table with the same columns as the input, but with the missing attributes filled in. Do not return any text other than the table. Mark filled attributes in bold. Do not return the tool outputs directly, or your reasoning, but only interpret them and return only the final table.
{% endmessage %}""",
    ),
)

pipeline.connect("summarizer.replies", "agent.messages")

output = pipeline.run({
    "fetcher": {
        "urls": [
            "https://www.booking.com/searchresults.html?ss=Berlin%2C+Berlin+Federal+State%2C+Germany&efdco=1&label=gen173nr-10CAEoggI46AdIM1gEaLYBiAEBmAEzuAEHyAEM2AED6AEB-AEBiAIBqAIBuAKQgrTOBsACAdICJDllMDgyNTA1LTdlYmMtNDJkNi04NzFmLWUyYTRkNWM1NjM3ZtgCAeACAQ&aid=304142&lang=en-us&sb=1&src_elem=sb&src=index&dest_id=-1746443&dest_type=city&ac_position=0&ac_click_type=b&ac_langcode=xu&ac_suggestion_list_length=5&search_selected=true&search_pageview_id=0da2508885520c7d&ac_meta=GhAwZGEyNTA4ODg1NTIwYzdkIAAoATICeHU6BmJlcmxpbg%3D%3D&checkin=2026-05-05&checkout=2026-05-08&group_adults=1&no_rooms=1&group_children=0",
        ]
    }
})

# Do not display the entire response, but only extract
# the table after filling the gaps.
response = output["agent"]["last_message"].text
table = response[response.index("| Hotel Name |"):]
display(Markdown(table))