"""MCP server exposing the tool layer over the Model Context Protocol.

Run standalone:  python -m src.tools.mcp_server
Any MCP-capable client (Claude Desktop, LangChain MCP adapters, etc.) can
then call the ratio engine, filings search, and currency conversion.

The same underlying functions are also exposed in-process as LangChain
tools (src/tools/langchain_tools.py) so the Gradio app works without a
separate server process.
"""

from __future__ import annotations

from mcp.server.fastmcp import FastMCP

from src.tools import external, ratios

mcp = FastMCP("financial-analyst-tools")


@mcp.tool()
def calculate_ratio(name: str, inputs: dict[str, float]) -> dict:
    """Compute a financial ratio. `name` is one of: roe, roa, ebitda_margin,
    current_ratio, quick_ratio, debt_to_equity, interest_coverage,
    free_cash_flow. `inputs` maps the formula's argument names to figures.
    """
    return ratios.compute(name, **inputs).as_dict()


@mcp.tool()
def calculator(expression: str) -> float:
    """Evaluate a plain arithmetic expression, e.g. '(1200 - 950) / 950'."""
    allowed = set("0123456789.+-*/() eE")
    if not set(expression) <= allowed:
        raise ValueError("Only arithmetic characters are allowed")
    return float(eval(expression, {"__builtins__": {}}, {}))  # noqa: S307 — charset-restricted


@mcp.tool()
def sec_edgar_search(company: str, form_type: str = "10-K") -> dict:
    """Search recent SEC EDGAR filings for a company."""
    return external.sec_edgar_search(company, form_type)


@mcp.tool()
def companies_house_search(company: str) -> dict:
    """Search UK Companies House for a company's registration details."""
    return external.companies_house_search(company)


@mcp.tool()
def convert_currency(amount: float, from_currency: str, to_currency: str) -> dict:
    """Convert an amount between currencies at current ECB rates."""
    return external.convert_currency(amount, from_currency, to_currency)


@mcp.tool()
def convert_file_to_markdown(path: str) -> str:
    """Convert a financial document (PDF, CSV, XLSX/XLS, JSON, PNG/JPG via OCR,
    or plain text) into clean Markdown with intact tables — the normalised form
    used for embedding and analysis."""
    from src.ingestion.loader import to_markdown
    return to_markdown(path)


if __name__ == "__main__":
    mcp.run()