FMP Data

Access financial market data through natural language queries.

Overview

The FMP (Financial Modeling Prep) LangChain integration provides a seamless way to access financial market data through natural language queries. This integration offers two main components:

• FMPDataToolkit: Creates collections of tools based on natural language queries
• FMPDataTool: A single unified tool that automatically selects and uses the appropriate endpoints

The integration leverages LangChain’s semantic search capabilities to match user queries with the most relevant FMP API endpoints, making financial data access more intuitive and efficient.

Setup

!pip install -U langchain-fmp-data

import os

# Replace with your actual API keys
os.environ["FMP_API_KEY"] = "your-fmp-api-key"  # pragma: allowlist secret
os.environ["OPENAI_API_KEY"] = "your-openai-api-key"  # pragma: allowlist secret

It’s also helpful (but not needed) to set up LangSmith for best-in-class observability:

# os.environ["LANGSMITH_TRACING"] = "true"
# os.environ["LANGSMITH_API_KEY"] = getpass.getpass()

Instantiation

There are two main ways to instantiate the FMP LangChain integration:

1. Using FMPDataToolkit

from langchain_fmp_data import FMPDataToolkit

query = "Get stock market prices and technical indicators"
# Basic instantiation
toolkit = FMPDataToolkit(query=query)

# Instantiation with specific query focus
market_toolkit = FMPDataToolkit(
    query=query,
    num_results=5,
)

# Instantiation with custom configuration
custom_toolkit = FMPDataToolkit(
    query="Financial analysis",
    num_results=3,
    similarity_threshold=0.4,
    cache_dir="/custom/cache/path",
)

2. Using FMPDataTool

from langchain_fmp_data import FMPDataTool
from langchain_fmp_data.tools import ResponseFormat

# Basic instantiation
tool = FMPDataTool()

# Advanced instantiation with custom settings
advanced_tool = FMPDataTool(
    max_iterations=50,
    temperature=0.2,
)

Invocation

The tools can be invoked in several ways:

Direct Invocation

# Using FMPDataTool
tool_direct = FMPDataTool()

# Basic query
# fmt: off
result = tool.invoke({"query": "What's Apple's current stock price?"})
# fmt: on

# Advanced query with specific format
# fmt: off
detailed_result = tool_direct.invoke(
    {
        "query": "Compare Tesla and Ford's profit margins",
        "response_format": ResponseFormat.BOTH,
    }
)
# fmt: on

Using with LangChain Agents

from langchain.agents import AgentExecutor, create_openai_functions_agent
from langchain_openai import ChatOpenAI

# Setup
llm = ChatOpenAI(temperature=0)
toolkit = FMPDataToolkit(
    query="Stock analysis",
    num_results=3,
)
tools = toolkit.get_tools()

# Create agent
prompt = "You are a helpful assistant. Answer the user's questions based on the provided context."
agent = create_openai_functions_agent(llm, tools, prompt)
agent_executor = AgentExecutor(
    agent=agent,
    tools=tools,
)

# Run query
# fmt: off
response = agent_executor.invoke({"input": "What's the PE ratio of Microsoft?"})
# fmt: on

Advanced Usage

You can customize the tool’s behavior:

# Initialize with custom settings
advanced_tool = FMPDataTool(
    max_iterations=50,  # Increase max iterations for complex queries
    temperature=0.2,  # Adjust temperature for more/less focused responses
)

# Example of a complex multi-part analysis
query = """
Analyze Apple's financial health by:
1. Examining current ratios and debt levels
2. Comparing profit margins to industry average
3. Looking at cash flow trends
4. Assessing growth metrics
"""
# fmt: off
response = advanced_tool.invoke(
    {
        "query": query,
        "response_format": ResponseFormat.BOTH}
)
# fmt: on
print("Detailed Financial Analysis:")
print(response)

Chaining

You can chain the tool similar to other tools simply by creating a chain with desired model.

from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI

# Setup
llm = ChatOpenAI(temperature=0)
toolkit = FMPDataToolkit(query="Stock analysis", num_results=3)
tools = toolkit.get_tools()

llm_with_tools = llm.bind(functions=tools)
output_parser = StrOutputParser()
# Create chain
runner = llm_with_tools | output_parser

# Run chain
# fmt: off
response = runner.invoke(
    {
        "input": "What's the PE ratio of Microsoft?"
    }
)
# fmt: on

API reference

FMPDataToolkit

Main class for creating collections of FMP API tools:

from typing import Any

from langchain.tools import Tool


class FMPDataToolkit:
    """Creates a collection of FMP data tools based on queries."""

    def __init__(
        self,
        query: str | None = None,
        num_results: int = 3,
        similarity_threshold: float = 0.3,
        cache_dir: str | None = None,
    ): ...

    def get_tools(self) -> list[Tool]:
        """Returns a list of relevant FMP API tools based on the query."""
        ...

FMPDataTool

Unified tool that automatically selects appropriate FMP endpoints:

# fmt: off
class FMPDataTool:
    """Single unified tool for accessing FMP data through natural language."""

    def __init__(
            self,
            max_iterations: int = 3,
            temperature: float = 0.0,
    ): ...

    def invoke(
            self,
            input: dict[str, Any],
    ) -> str | dict[str, Any]:
        """Execute a natural language query against FMP API."""
        ...

# fmt: on

ResponseFormat

Enum for controlling response format:

from enum import Enum


class ResponseFormat(str, Enum):
    RAW = "raw"  # Raw API response
    ANALYSIS = "text"  # Natural language analysis
    BOTH = "both"  # Both raw data and analysis